data.json

{
  "projects": [
    {
      "student_name": "Sophia G.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/25fb3536164ff3412efca7e41e08e85763b9c40f.png",
      "domains": [
        "Humanitarian Efforts and Social Justice",
        "Published Papers and Science Fairs"
      ],
      "project_title": "A Machine Learning Approach to Understanding the Determining Factors of the Gender Wage Gap",
      "project_desc": "Gender inequality is a complex subject consisting of a variety of issues and nuances. In this project, we choose to study gender income inequality—a prevalent issue in current society. Among the many factors that play a role in the gender wage gap, we focus on the affects of marital status, race, geographical location (by state), age, and years of education. By using these variables to create a model able to predict the hourly wage gap between a woman and their equivalent male counterpart, we can analyze the impact of each variable to better understand the role they play in the income gap. Utilizing income data from the Current Population Survey, we train and test five models—a Linear Regression, Decision Tree Regressor, Random Forest Regressor, KNeighbors Regressor, and MLP Regressor. Our Linear Regression model found that there is a correlation between being a never married worker and a smaller gender wage gap, as well as being a married worker with an absent spouse and a greater gender wage gap. In general, though, our models found little correlation between the variables provided and the predicted hourly age gap. \n",
      "research_paper": "assets/pdfs/4c2dec4fb3c31bc305c9008eb6addb04d1747d92.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/sophiaguan/genderwagegapanalysis",
      "graphic_link": "http://media.wbur.org/wp/2016/04/0416_oag-pay-21-1000x666.jpg",
      "headline": " By studying the affect of different attributes on the gender wage gap, we can better understand both the scale of this issue and its possible solutions. So, we explore the question, how does a worker’s marital status, along with other variables, impact the gap in hourly wage between male and female workers? We seek to create a model able to predict the gender wage gap given a set of variables—age, years of education, race, state, and marital status.",
      "project_id": "0a574bde2fff76c3dafb371ae2cef1cfbd7ed854",
      "expand": true,
      "tags": [
        "Wage Gap",
        "Economics",
        "Published",
        "Science Fairs"
      ],
      "published": true,
      "science_fairs": "2nd Place at San Diego BROADCOM Science Fair (Senior Division)",
      "related_proj": [
        21,
        133,
        7
      ]
    },
    {
      "student_name": "Alice H.",
      "mentor_name": "Tony Rodriguez",
      "mentor_title": "PhD Candidate in Astrophysics at Caltech",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/e3c71b4cb5bdd5775de029db904c939030e0e618.png",
      "domains": [
        "Physics",
        "Published Papers and Science Fairs"
      ],
      "project_title": "A Hybrid CNN-LSTM Model For Predicting Solar Cycle 25",
      "project_desc": "The solar cycle is linked to the number of visible sunspots and follows the fluctuations of the Sun’s magnetic field. It can have powerful global impacts on the Earth. Thus, predicting the timing and amplitude of the peak of the incoming solar cycle 25 is of great importance. This study uses a hybrid deep learning CNN-LSTM model and observed 13-month smoothed sunspot numbers to predict Solar Cycle 25. The MinMax normalization substantially reduces the error of the CNN-LSTM model’s solar cycle predictions compared to the Standard Deviation normalization. The results suggest that it is best to use four historical solar cycles to predict the future solar cycle. The predicted Solar Cycle 25 will have a peak amplitude similar to the 13-month smoothed peak of solar cycle 24. The predicted solar cycle 25 peak spans a relatively long period of time between approximately August 2023 and July 2024. ",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://media.wtsp.com/assets/WTSP/images/20519ffa-0230-435b-8f2b-46718cd853c8/20519ffa-0230-435b-8f2b-46718cd853c8_750x422.jpg",
      "headline": "The goal of this study is to predict Solar Cycle 25 through the deep learning approach,and determine what parameters affect prediction accuracy and what the optimal number of historical solar cycles are used to reliably and accurately predict the upcoming solar cycle. The solar cycle predictions will help us prepare ahead of time for future solar activity.",
      "project_id": "073776ccdb02a715b1a9c4ea590c72703600cadc",
      "expand": false,
      "tags": [
        "LSTMs",
        "Solar Cycles",
        "Astronomy",
        "Published"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "publication_link": "https://www.jsr.org/hs/index.php/path/article/view/3996",
      "related_proj": [
        36,
        12,
        2
      ]
    },
    {
      "student_name": "Alex M.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Business and Finance",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Applications of AI in Microfinance ",
      "project_desc": "Using AI models to predict who will be able to get a microloan",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://akm-img-a-in.tosshub.com/businesstoday/images/story/202007/investment_660x450_190220100047_200720121651.jpg",
      "headline": "I hope to explore how to use AI in the field of microfinance to help reduce income inequality. Microfinance has greatly helped decrease rural poverty rates in Bangladesh. The leader of this effort won the Nobel Peace Prize for his work. These micro-loans give opportunity to those who are not otherwise able to obtain financing for their entrepreneurial ideas. This can be applied in the U.S. too, in places where the population cannot otherwise obtain loans to start small businesses and climb their way out of poverty. It can be very difficult to start from rock bottom in the US, especially for those without access to resources. If people could obtain small loans and start small businesses, they could work their way out of poverty. The microloans have to be viable for banks as well. The problem I’d like to explore is whether AI/ML can be used to determine how to deploy microloans efficiently to address income inequality in the U.S.",
      "project_id": "aeb3a7e9bdee8ba99533d1e263521562419a8fb6",
      "expand": false,
      "tags": [
        "Economics",
        "Microfinance",
        "Inequality",
        "Published"
      ],
      "published": true,
      "science_fairs": "Massachusetts Science and Engineering Fair (MSEF)",
      "related_proj": [
        5,
        26,
        186
      ]
    },
    {
      "student_name": "Roberto T.",
      "mentor_name": "Sophia Barton",
      "mentor_title": "Computer Science MS from Stanford",
      "mentor_image": "assets/images/mentor_imgs/56c2b2fb8dc613f5a0a7ae222a5ee595734e2950.png",
      "student_image": "assets/images/student_imgs/3794c48b2334ab46228d00da2c6c7e612e5bc9e9.png",
      "domains": [
        "Physics",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Stellar Classification based on Numerous Characteristics using Machine Learning",
      "project_desc": "The task of stellar classification can be tedious and\nlengthy when done manually. One can expedite stellar classifi-\ncation by creating an artificial intelligence model to automate\nthe process. As we as a species continue to explore the frontier\nof the observable universe, we should seek to automate time\nintensive problems like stellar classification. The current stellar\nclassification model serves to effectively categorize stars for re-\nsearch purposes regarding their distribution around the universe,\nso automating the development of this resource would allow\nprofessionals to allocate more time to explore the bounds of our\ncurrent understanding of space and the universe. After finding\nand analyzing a dataset containing numerical and categorical\nfeatures, a supervised learning approach was then used to train\nand test different models on their ability to classify the stars\nin the given test set. A Decision Tree Classifier, Random Forest\nClassifier, Ridge Classifier, and Support Vector Classifier were\ntrained and tested using the data. The most successful models\nwere the Decision Tree Classifier and Random Forest Classifier,\neach with about a 94 percent prediction accuracy across different\naccuracy metrics on the test data. Despite some drawbacks in\nregards to the availability of usable data, four models were\ntrained and two were proven to be consistently and successfully\naccurate. Any future attempts at developing models for stellar\nclassification should concentrate more on gathering data as to\nhave a more thoroughly trained set of models.",
      "research_paper": "assets/pdfs/7b342766e5c2f5ed3fcf18d4910c0e3513323044.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://www.universetoday.com/wp-content/uploads/2008/11/tarantula.jpg",
      "headline": "The task of stellar classification can be tedious and lengthy when done manually. One can expedite stellar classification by creating an artificial intelligence model to automate the process. The current stellar classification model serves to effectively categorize stars for research purposes regarding their distribution around the universe, so automating the development of this resource would allow professionals to allocate more time to explore the bounds of our current understanding of space and the universe. After finding and analyzing a dataset containing numerical and categorical features, a supervised learning approach was then used to train and test different models on their ability to classify the stars in the given test set. A Decision Tree Classifier, Random Forest Classifier, Ridge Classifier, and Support Vector Classifier were trained and tested using the data. ",
      "project_id": "5ee1177c671abf3e11897fc2ebbba3e1355573d4",
      "expand": true,
      "tags": [
        "Astronomy",
        "Stellar Classification",
        "Published"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "publication_link": "https://www.jsr.org/hs/index.php/path/article/view/4375",
      "related_proj": [
        60,
        18,
        36
      ]
    },
    {
      "student_name": "Elgin V.",
      "mentor_name": "Joseph Vincent",
      "mentor_title": "Aerospace Engineering PhD Candidate at Stanford",
      "mentor_image": "assets/images/mentor_imgs/808f2a4f0dad138963d86e72fa8a6bca1820e194.png",
      "student_image": "assets/images/student_imgs/235e3e4102d4ee2526e5af8047f3ed72e18f083f.png",
      "domains": [
        "Sports and Motion Sciences",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Predicting Running Injuries with Machine Learning Models",
      "project_desc": "Is it possible to predict running injuries with only a dataset and machine learning models? This paper explores this question by using classification models, including the Logistic Regression model and the Random Forest Classifier model. In the dataset used, ten features were taken into account when predicting running injuries. With slight modifications, the Weighted Logistic Regression and over and down-sampling Random Forest Classifier models were used to mitigate the imbalance in the dataset. The results suggested that the best model was Weighted Logistic Regression and that the best score metric to take into account was the F beta score.",
      "research_paper": "assets/pdfs/be88cd8430de161d40c89667573175ca23a8a90e.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.runtastic.com/blog/wp-content/uploads/2021/05/thumbnail_1200x800-2.jpg",
      "headline": "Is it possible to predict running injuries with only a dataset and machine learning models? This paper explores this question by using classification models, including the Logistic Regression model and the Random Forest Classifier model.",
      "project_id": "8f11c9cea7ec6c6ef1b452880aceade84635e612",
      "expand": false,
      "tags": [
        "Sports",
        "Running",
        "Logistic Regression"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "publication_link": "https://www.jsr.org/hs/index.php/path/article/view/4046",
      "related_proj": [
        13,
        15,
        0
      ]
    },
    {
      "student_name": "Rayyan M.",
      "mentor_name": "Philip Bell",
      "mentor_title": "Computer Science MS from Georgia Tech, Harvard and Oxford alum, AI Engineer at Cognino ",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/eef1734ba79e5067353abc408ba77616a654a8e7.png",
      "domains": [
        "Environment",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Combating Climate Fake News Using NLP",
      "project_desc": "As fake news becomes more prevalent across the US, important issues become harder to solve. One such issue is climate change, where climate misinformation has worsened viewer’s abilities to distinguish between fake information and real information. This project’s objective is to tackle climate misinformation using an artificial intelligence model. The model utilizes a BERT model tested on the “climate_fever” dataset to classify whether climate-related claims are true based on pieces of evidence.",
      "research_paper": "assets/pdfs/d80b43b85a066c44470644a3696aca2d9ecfac57.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/SpicyTaco17/Climate-Fake-News",
      "project_webpage": "https://spicytaco17-climate-fake-news-app-57cfkv.streamlit.app/",
      "graphic_link": "https://media-cldnry.s-nbcnews.com/image/upload/newscms/2018_25/2470661/189619-weather-forecasting-mn-1620.jpg",
      "headline": "As fake news becomes more prevalent across the US, important issues become harder to solve. One such issue is climate change, where climate misinformation has worsened viewer’s abilities to distinguish between fake information and real information. This project’s objective is to tackle climate misinformation using an artificial intelligence model.",
      "project_id": "1dbfe39aa305fa7f40158cb17cd70a167c56d275",
      "expand": false,
      "tags": [
        "News",
        "Fake News",
        "Climate Change",
        "Natural Language Processing"
      ],
      "published": true,
      "science_fairs": "Santa Clara ISEF Qualifier",
      "related_proj": [
        27,
        67,
        2
      ]
    },
    {
      "student_name": "Arnav D.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "The Differentiation of Viral and Bacterial Pneumonia using Deep Learning",
      "project_desc": "This project aims to find out whether a Convolutional Neural Network can be used to classify x-ray scans as having either bacterial or viral Pneumonia. First, its use is discussed, as it is meant to be used to assist with pulmonary healthcare in underprivileged areas. Then, its structure is outlined as a CNN set up using Keras and Tensorflow that is trained on a dataset consisting of 5856 x-ray scans of patients one to five years old from Guangzhou Women and Children’s Medical Center. After testing multiple setups, the final model, based on Inception ResNet v2, differentiates between the two types of Pneumonia with a 75% test accuracy, using minimal epochs in order to increase efficiency. The model was further improved by using dropout, the implementation of early stopping, L2 regularization and augmentation of the training data in order to combat overfitting. Lastly, the model’s limitations and future are discussed. This model proves one can predict and identify types of Pneumonia using a CNN with relatively high accuracy and shows the potential of deep learning in the context of medical imaging to assist medical staff.",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://thorax.bmj.com/content/thoraxjnl/57/5/438/F1.large.jpg",
      "headline": "This project aims to find out whether a Convolutional Neural Network can be used to classify x-ray scans as having either bacterial or viral Pneumonia. ",
      "project_id": "58dec3622ea48a99256e07cb547de3f8418e0c94",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Pneumonia",
        "Classification",
        "Published"
      ],
      "published": true,
      "science_fairs": "2nd Place at Orange County Science Fair, California Science and Engineering Fair (CSEF)",
      "related_proj": [
        98,
        119,
        16
      ]
    },
    {
      "student_name": "Avyukth H.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/431caaf556b23766fb3909779a4d98cb8b62de72.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Impact of Class Weights and Feature Importance in Automated Stroke Detection",
      "project_desc": "A stroke occurs when a blood vessel that carries oxygen and nutrients to the brain is either blocked by a clot or ruptures. Consequently, part of the brain is unable to obtain the blood (and oxygen) it needs, so brain cells die. This makes it important to be able to assess the probability of a stroke given features that are specific to patients so that they could take preventative measures in the future. Thus, the unpredictability and deadliness of strokes pose the following questions: Can we predict the occurrence of a stroke from few facts about the patient easily accessible by the doctor? What are the most important features for stroke prediction? In this project, we investigate the feasibility of using a supervised machine learning model to predict stroke occurrence. In practice, however, we faced challenges such as low prevalence and the imbalance in the available dataset, with many more negative than positive cases. In this research paper, we do a parametric study of class weighting as a way to tackle imbalance during training. We then infer the most important features that should be taken into consideration for stroke prediction. Assessing feature importance allows for patients to focus on two or three areas that may be contributing to their high probability for getting a stroke. The most significant result was that the most important feature that should be considered when determining the probability of an individual getting a stroke is age. However, there is no distinct second most important feature. Additionally, a non monotonic improvement with a class weight of 22.5 for positive cases in this dataset produces the most optimal results. \n",
      "research_paper": "assets/pdfs/070cedde410910544ceefd49565631e69a685b9a.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.medrxiv.org/content/medrxiv/early/2021/04/07/2021.04.02.21253083/F1.large.jpg",
      "headline": "In this research paper, we do a parametric study of class weighting as a way to tackle imbalance during training. We then infer the most important features that should be taken into consideration for stroke prediction. ",
      "project_id": "c5872d603a0e4ff94f5ecab64817175d43b9961c",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Strokes",
        "Neuroscience"
      ],
      "published": true,
      "publications": "Curieux Academic Journal",
      "related_proj": [
        67,
        2,
        19
      ]
    },
    {
      "student_name": "Avnith V.",
      "mentor_name": "Jacklyn Luu",
      "mentor_title": "Biomedical Informatics MS at Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Evaluating Machine Learning Models on Predicting Change in Enzyme Thermostability",
      "project_desc": "Enzymes are efficient catalysts for biological reactions, and can potentially be designed to speed up non-biological reactions, such as reactions in industrial processes. However, physically experimenting with new protein designs is time consuming, and an efficient method to predict protein stability is needed. Our research problem is finding the best machine learning model to predict the change in enzyme thermostability after a single point mutation in the amino acid sequence. We trained several machine learning models and found that the XGBoost model had the best performance with an R2 score of 0.593 (R2 score is a metric where higher is better and a perfect model would have a score of 1). ",
      "research_paper": "assets/pdfs/2534c03998d293d48a8b45f9a7341724d2d1e3df.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "github": "https://github.com/avnithv/enzyme-thermostability-project",
      "graphic_link": "https://www.innovationnewsnetwork.com/wp-content/uploads/2020/06/cold-adapted-enzymes-696x392.jpg",
      "headline": "Our research problem is finding the best machine learning model to predict the change in enzyme thermostability after a single point mutation in the amino acid sequence. ",
      "project_id": "c6626d41c51d0fda47297103ef1343bd58a9b435",
      "expand": true,
      "tags": [
        "Biology",
        "Chemistry",
        "DNA",
        "Gradient Boosting"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "publication_link": "https://www.jsr.org/hs/index.php/path/article/view/4364",
      "related_proj": [
        2,
        21,
        73
      ]
    },
    {
      "student_name": "Anjali S.",
      "mentor_name": "Katie O'Nell",
      "mentor_title": "PhD Student at Dartmouth, Brain and Cognitive Sciences BS from MIT, Ethics+Social Sciences project and course developer at Inspirit AI",
      "mentor_image": "assets/images/mentor_imgs/3df727e01bd6076d80bdefdd1c47288c998e0f7e.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Differences in predicted rates of vaginal births after cesarean across racial groups in a ‘race-neutral’ model",
      "project_desc": "When physicians and pregnant patients make decisions about whether to pursue a vaginal birth or cesarean, there are many factors at play. While vaginal birth can have health benefits for both parent and child, there are significant safety risks. In order to minimize these risks, physicians use predictive models to determine how likely patients are to have successful vaginal births after cesareans (VBAC). For many years, these predictive models included race as a variable. This decision recently came under fire, and the Maternal Fetal Medicine Unit (MFMU) published a calculator that did not include race as a variable, but still predicted VBAC success with high accuracy. A large body of work in machine learning has highlighted that supposedly de-biased systems often re-code sensitive variables like race in terms of proxy variables. In order to determine if this was the case in this calculator, we replicated their formula, then found base-rate statistics of all the input variables for three different racial groups: Black, White, and Asian. We found that the distribution of VBAC probabilities for our simulated patients from these three groups were, indeed, significantly different from each other. Further, the predicted VBAC rates increased as a function of societal marginalization: Black patients were 47.6% likely to have a successful VBAC, Asian patients had a 48.6% probability, and White patients had a 49.4% probability. While these values are all within a few percentage points of each other, the differences in these simulated distributions shows how there may still be underlying disparities in the maternal healthcare system.",
      "research_paper": "assets/pdfs/4a1e29bb400296758e8d804dbefc5ac88a058f7f.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.verywellfamily.com/thmb/m5vCHw2YW_zteiIwmYXmJ2QRosE=/1500x0/filters:no_upscale():max_bytes(150000):strip_icc()/AA6030-001-56a7713e3df78cf77295fbe7.jpg",
      "headline": "A large body of work in machine learning has highlighted that supposedly de-biased systems often re-code sensitive variables like race in terms of proxy variables. In order to determine if this was the case in this calculator, we replicated their formula, then found base-rate statistics of all the input variables for three different racial groups: Black, White, and Asian.",
      "project_id": "72bcaf1817b1e88e623d2b3f28f29d58dd65d048",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Women's Health",
        "Racial Equality",
        "Gender Equality"
      ],
      "published": true,
      "publications": "medRxiv Medical Ethics preprint",
      "publication_link": "https://www.medrxiv.org/content/10.1101/2023.10.14.23296978v1",
      "related_proj": [
        167,
        104,
        184
      ]
    },
    {
      "student_name": "Shanzeh H.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/79f656245a5e9bd9b5892bae9b8c3fb9ad33f352.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "The Utilization of Artificial Intelligence in Enabling the Early Detection of Brain Tumors",
      "project_desc": "Diagnosing brain tumors is challenging due to their location and varied presentations\nthat may mimic common disorders. A cancer diagnosis can be missed even when advanced imaging is conducted due to interpretive error or an incompatible clinical history as presented. Machine learning, when applied to radiological imagery, can alert physicians earlier to the presence of tumors and improve diagnostic evaluation. This enhanced evaluation can lead to earlier detection of malignant tumors and positively improve prognosis, quality of life, and treatment. This research aims to investigate the application of machine learning to enhance diagnosis. The study developed two machine learning models, a logistic regression model, and a neural network model. We hypothesized that our methods would work sufficiently, proving the correct diagnosis rate, especially within the neural networks model, as they are more complicated in nature. Applying a dataset sourced from Kaggle into the respective algorithms showcased a promising future of machine learning applications to brain tumor diagnosis, with test accuracies in the logistic regression model high (68%) and the neural network model at a significant high (84%). The simple fact of achieving a 84% accuracy rate on new data in the neural network model represents a promising future for the early detection of brain tumors.\n",
      "research_paper": "assets/pdfs/00fd300026214ae10e0b11d2cd3da2869f219dba.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://scopeblog.stanford.edu/wp-content/uploads/2022/10/Scope-Arami-Nanostars-1024x578.jpg",
      "headline": "This research aims to investigate the application of machine learning to enhance diagnosis.",
      "project_id": "0a26e050296441b0075ca224251dbbc48c833cb3",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Strokes",
        "Neuroscience",
        "Published",
        "Logistic Regression"
      ],
      "published": true,
      "publications": "Journal of Emerging Investigators",
      "related_proj": [
        133,
        131,
        135
      ]
    },
    {
      "student_name": "Ashray P.",
      "mentor_name": "Christopher Mauck",
      "mentor_title": "Computer Science Masters from MIT",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Sports and Motion Sciences",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Optimizing Prediction Accuracy Using Advanced Ensemble And Voting Classifier Methods",
      "project_desc": "Machine learning models have become more advanced over the last few years, allowing for increased prediction accuracy. One relatively undiscovered model-making technique is the Voting Classifier, which is able to combine the predictions of different ML models to arrive at one output prediction. Harnessing this method can allow for increased accuracy and revolutionize the future of Artificial Intelligence (AI) and Machine Learning (ML).\n\nThis project observes how various machine learning models, once tuned, can further be combined to create a complex model that uses NFL data from the past 18 years to predict the outcomes of matchups between any two competing teams. Through an extensive study of various existing models and key parameters that are critical to the sport of American football, a sophisticated advanced model was created that is highly sensitive to a handful of most important parameters in accurately predicting a given matchup at a level that consistently matches or beats Vegas predictions.",
      "research_paper": "assets/pdfs/b973c789e1df2e7826d3db14ebfcdcb3bffa19f9.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "project_webpage": "https://avp123-nflstreamlit-nflapp-z925xp.streamlit.app/",
      "graphic_link": "https://dims.apnews.com/dims4/default/53279da/2147483647/strip/false/crop/3000x2000+0+0/resize/1486x991!/quality/90/?url=https%3A%2F%2Fstorage.googleapis.com%2Fafs-prod%2Fmedia%2Fc94ccae5a2354fe3a486e891c2b781e4%2F3000.jpeg",
      "headline": "This project observes how various machine learning models, once tuned, can further be combined to create a complex model that uses NFL data from the past 18 years to predict the outcomes of matchups between any two competing teams.",
      "project_id": "f6c0acfb6db69f9cc850d744d56092222ac474e4",
      "expand": true,
      "tags": [
        "Sports",
        "Pop Culture",
        "Published",
        "Prediction"
      ],
      "published": true,
      "science_fairs": "2nd Place at Contra Costa Science and Engineering Fair",
      "related_proj": [
        22,
        16,
        8
      ]
    },
    {
      "student_name": "Armita K.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Diagnosing Brain Tumors from MRI Images Using Deep Transfer Learning",
      "project_desc": "Each year, more than 100,000 people in the United States are diagnosed with a brain tumor. An early and accurate diagnosis is crucial in getting patients the necessary treatment and increasing survival rates. In recent years, machine learning algorithms have become increasingly popular in the medical field due to their ability to recognize complex patterns and reduce human errors. However, accurate diagnosis using deep learning algorithms requires a large amount of training data, which is not always available. Additionally, training a model from scratch can take a long time and requires vast amounts of computational power. As a solution, this study aims to utilize a transfer learning method in which the prior knowledge of a pretrained model is used to aid in a new classification problem. In this study, a dataset of MRI images consisting of four classes (no tumor, pituitary tumor, meningioma, and glioma) were used. The performance of seven pretrained models (ResNet18, ResNet50, VGG16, DenseNet, GoogLeNet, ShuffleNet, and MobileNet) were evaluated in order to see which would achieve the highest classification accuracy. Additionally, this study examined two different methods for the implementation of transfer learning. In the first method, all layers of the pretrained model were frozen and in the second method, all layers of the pretrained model were trained. The best performing models proved to be ResNet18 and ShuffleNet with all layers trained, achieving an accuracy of 97.86%. The results also showed that the unfrozen models outperformed their frozen counterparts.  \n",
      "research_paper": "assets/pdfs/e1797a211b5825afc3630eb6b81d399b57f31911.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://assets.medpagetoday.net/media/images/99xxx/99832.jpg?width=0.6",
      "headline": "This study aims to utilize a transfer learning method in which the prior knowledge of a pretrained model is used to aid in a new classification problem.",
      "project_id": "d50b4f55ed899b280df774d50ecd798c54043892",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Strokes",
        "Neuroscience",
        "Published"
      ],
      "published": true,
      "publications": "Journal of High School Research",
      "publication_link": "https://jhss.scholasticahq.com/article/89930",
      "related_proj": [
        3,
        68,
        133
      ]
    },
    {
      "student_name": "Dhruv M.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Diversified AI Techniques for Augmenting Brain Tumor Diagnosis",
      "project_desc": "Brain tumors affect thousands of people worldwide each year and can be extremely fatal if not diagnosed early They are challenging to diagnose due to their complexity and the overlapping features of different tumor types. This research explores the application of AI technology to expedite the diagnosis of brain tumors. The proposed AI-based approaches involve using deep learning algorithms to analyze medical imaging data, specifically MRI scans. The goal was to build a robust and accurate model that could overcome distribution shifts. Some of the models used include classical machine learning models and a convolutional neural network. The results demonstrate that AI-based approaches can significantly improve the accuracy and expedite the process of brain tumor diagnosis. The performances of the models were evaluated by using cross-validation and measuring accuracy, using a completely different dataset of MRI scans, to assess how the models performed when dealing with distribution shifts. The logistic regression model achieved a testing accuracy of 78.56%. The multi-layer perceptron (MLP) model achieved a testing accuracy of 74.89%. The multiplicative weight update method combined two models\n(MLP and logistic regression) with dynamically adjusted weights and achieved a testing accuracy of 83.83%. An approach where multiple aggregating models were used collaboratively achieved a testing accuracy of 86.46%. The best performing model, a convolutional neural network (CNN), yielded a testing accuracy of 98.20%.",
      "research_paper": "assets/pdfs/a1958ab872b0c13f190ad9056b3cdc9807b3356e.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "github": "https://github.com/hello11210/AI-for-Brain-Tumor-Diagnosis",
      "graphic_link": "https://cdn.images.express.co.uk/img/dynamic/11/590x/1682717_1.jpg",
      "headline": "This research explores the application of AI technology to expedite the diagnosis of brain tumors.",
      "project_id": "cc63fba2402223c028df68372b1779aa8e947df6",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Strokes",
        "Neuroscience",
        "Published"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "related_proj": [
        101,
        133,
        185
      ]
    },
    {
      "student_name": "Surya K.",
      "mentor_name": "Sriram Hathwar",
      "mentor_title": "Princeton Alum, Bioinformatics Researcher",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/e882056b2810f35a11fd1195b408148cdebea4d8.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Diagnosing Hypertrophic Cardiomyopathy Using Machine Learning Models on CMRs and EKGs of the Heart",
      "project_desc": "Hypertrophic cardiomyopathy (HCM) is a common inherited heart disorder manifesting as hypertrophy of the left ventricle of the heart.  However, it often goes undiagnosed, which we must seek to avoid since the possibility of sudden cardiac death (SCD) as a result of HCM is not insignificant.  In this project, we presented a pair of models, one CNN model and one Long Short Term Memory (LSTM) model, that are capable of classifying cardiac magnetic resonance (CMR) and heart electrocardiogram (EKG) scans, respectively. We hypothesized that machine learning techniques applied to CMR and EKG data can predict HCM with a high degree of accuracy, precision, recall, and F1 Score.  Each of these models classifies their respective scans into HCM and non-HCM categories.  The CNN model has an accuracy of 94.71%, a precision of 96.97%, a recall of 91.21%, and an F1 score of 94.85%.  The LSTM model has an accuracy of 90.51%, a precision of 60.31%, a recall of 60.08%, and an F1 score of 60.19%.  These results show that these machine learning models are viable tools that could assist physicians in the diagnosis of HCM patients.",
      "research_paper": "assets/pdfs/e305dc45b1659fe5cba7c86b83075c46b1023f9f.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/suryakolluri6/HCM_diagnostic_models",
      "graphic_link": "https://www.ucl.ac.uk/news/sites/news/files/heart_stock_cropped_0.jpg",
      "headline": "In this project, we presented a pair of models, one CNN model and one Long Short Term Memory (LSTM) model, that are capable of classifying cardiac magnetic resonance (CMR) and heart electrocardiogram (EKG) scans, respectively.",
      "project_id": "3d64a57084fa3a9804772cbb3d3518cebdd44bea",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Heart Disease",
        "Published"
      ],
      "published": true,
      "publications": "Journal of Emerging Investigators",
      "related_proj": [
        34,
        42,
        108
      ]
    },
    {
      "student_name": "Katarina A.",
      "mentor_name": "Ayush Pandit",
      "mentor_title": "PhD Candidate at Stanford, prior Stanford Bioengineering Alum",
      "mentor_image": "assets/images/mentor_imgs/8ebea060f7acf53ebda820482eaa4ee26d37df4b.png",
      "student_image": "assets/images/student_imgs/16e627973984da51e6911c2796361a48a3207583.png",
      "domains": [
        "Environment",
        "Published Papers and Science Fairs"
      ],
      "project_title": "AI-Based Image Classification Used to Accurately Distinguish Recyclable Material Versus Non-Recyclable Material",
      "project_desc": "In today’s world, pollution is increasing as plastics and other materials are not recycled properly, resulting in landfills. One cause of this improper disposal of materials is that it can be difficult to tell if a material is able to be recycled. In response, I created a machine learning model that can distinguish recyclable materials from trash through image classification. For my model, I used a dataset called trash-net. I first extracted the contents of the data and resized the dataset in order to have better organization. There are six categories within the dataset: cardboard, glass, metal, paper, plastic, and trash, that the images are organized in. I used resnet34 which is a pre-trained convolutional neural network (CNN) in order to perform the image classification. Afterwards, I trained my model by running the program repeatedly and then tested it by seeing if it accurately predicted if a material was recyclable or not. Lastly, I used matplotlib to visualize the results. The accuracy of the model ended up being about 88%. Generally, if the accuracy of a machine-learning model is higher than 50% then it performs relatively well. With more training and a greater number of images, the program could potentially increase in accuracy. In conclusion, I think my model would help as it could generally classify if a material was recyclable or not. However, an application in which the user directly scans an item would be more useful. Nevertheless, my model performed well and enabled me to learn more about the use of artificial intelligence. \n\n",
      "research_paper": "assets/pdfs/0f175d18ad7de4d88554f88bca3a0959c5b2c2fc.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://s3-prod.crainsnewyork.com/s3fs-public/REAL_ESTATE_130729880_AR_-1_0.jpg",
      "headline": "One cause of this improper disposal of materials is that it can be difficult to tell if a material is able to be recycled. In response, I created a machine learning model that can distinguish recyclable materials from trash through image classification.",
      "project_id": "3dd373e176952ff4830b8d1d532e0973054b9d40",
      "expand": true,
      "tags": [
        "Environmentalism",
        "Recycling",
        "Sustainability"
      ],
      "published": true,
      "science_fairs": "Synopsys Science Fair",
      "related_proj": [
        11,
        21,
        120
      ]
    },
    {
      "student_name": "Jack L.",
      "mentor_name": "Peter Washington",
      "mentor_title": "Computer Science Professor at UHawaii, CS PhD from Stanford",
      "mentor_image": "assets/images/mentor_imgs/19db8abe6f5ba342397cabd447a93cbb5d008688.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Psychology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Investigating Data Augmentation Strategies for Computer Vision Facial Expression Recognition",
      "project_desc": "Autism is a neurodevelopmental disorder. A major symptom is difficulty communicating and understanding social cues such as emotions. I aim to help people with autism better recognize emotions by developing improved artificial intelligence (AI) models to recognize facial expressions. Such models can be and have been integrated into digital therapeutics for children with autism. A crucial step to achieving performant models is to apply data augmentation to increase the dataset size and the generalization capacity. I compare and contrast data augmentation strategies on the Facial Expression Recognition (FER) 2013 dataset to determine which method leads to a maximal increase in performance. I then examine the benefit of data augmentation at various training set sizes. Among the strategies I evaluate, I find that shifting the width of the image provides the greatest increase to performance when compared to not applying data augmentation. Furthermore, I find that at several training dataset sizes ranging from 100 to 20,000 images, applying all data augmentation strategies consistently outperforms no data augmentation. These strategies can inform the development of digital therapies for autism which focus on the evocation and subsequent automatic detection of facial expressions.",
      "research_paper": "assets/pdfs/16737c9b12a4149b57d291e8d624dca57062a193.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://149695847.v2.pressablecdn.com/wp-content/uploads/2020/04/Learn-Facial-Recognition-scaled.jpg",
      "headline": "I aim to help people with autism better recognize emotions by developing improved artificial intelligence (AI) models to recognize facial expressions.",
      "project_id": "2157806554f790856dcb6cc95179a733d8161679",
      "expand": true,
      "tags": [
        "Computer Vision",
        "Facial Emotion",
        "Developmental Disorders"
      ],
      "published": true,
      "publications": "Curieux Academic Journal",
      "publication_link": "https://www.curieuxacademicjournal.com/_files/ugd/99711c_cae123aa84284189b8142087c8589d21.pdf",
      "related_proj": [
        15,
        40,
        103
      ]
    },
    {
      "student_name": "Varsha N.",
      "mentor_name": "Roger Jin",
      "mentor_title": "Computer Science MS at MIT, ML researcher at Apple",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "A Novel Approach to Promote Equity in Skin Disease Diagnosis by AI Models",
      "project_desc": "AI-based systems are increasingly used to diagnose skin diseases with datasets available on the internet. However, the training data predominantly represents fair skinned people. The goal is to increase the accuracy of existing models in diagnosing skin disease across various skin tones within 10% of that obtained in diagnosing fairer skin tones, which is about 95.8%. The publicly available HAM10000 dataset with 10000 fair skin images was preprocessed and used to fine-tune a Vision Transformer model trained on the ImageNet-21K dataset. The model obtained about 94.9% accuracy in diagnosing fair skin images and about 19.8% accuracy in diagnosing real non-fair skin images. Next, 50% of the training data was transformed into non-fair skin simulation by CycleGAN based on 320 real non-fair images. After several iterations, the model achieved about 78.6% accuracy in classifying any skin tone and 52.1% accuracy in classifying real non-fair images. Though this 68.4% is outside of the desired range, it still falls within the accuracy of dermatologists (48–77%). However, it is the accuracy in classifying non-fair skin that makes this model significant. The model achieved a 28.6% increase in accuracy in diagnosing non-fair images using simulated data thereby narrowing the disparity. A confusion matrix was plotted to visualize the validity of the predictions. The functionality of this model suggests that similar data augmentation techniques could be applied to other AI models to ensure their fairness to all categories of people and correct any biases in data due to historic under-representation.",
      "research_paper": "assets/pdfs/2ee4416a991da2d5b43d584262c76998e321b45f.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.vchri.ca/sites/default/files/styles/max_1300x1300/public/field/image/news_20180724_sunsray.jpg?itok=gFON3EMy",
      "headline": "The goal is to increase the accuracy of existing models in diagnosing skin disease across various skin tones within 10% of that obtained in diagnosing fairer skin tones, which is about 95.8%.",
      "project_id": "9941206fc0938ca237fdb718341d95aff57a5192",
      "expand": false,
      "tags": [
        "Skin Cancer",
        "Diseases and Health Conditions",
        "Diseases",
        "Science Fairs"
      ],
      "published": true,
      "science_fairs": "3rd Place in the Alameda County Science Fair",
      "related_proj": [
        29,
        133,
        9
      ]
    },
    {
      "student_name": "Cherry (.",
      "mentor_name": "Shreya Parchure",
      "mentor_title": "MD-PhD student at UPenn",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Machine Learning Approaches to Detect Brain Tumors from Magnetic Resonance Imaging Scans",
      "project_desc": "Artificial intelligence (AI) models have brought about a significant transformation in various industries, including healthcare, in recent years. Among the many areas benefiting from AI, brain tumor detection has seen remarkable advancements. Accurate brain tumor detection plays a crucial role in the timely diagnosis and treatment of neurological disorders. Thanks to AI models, the process of detecting brain tumors has become more precise and efficient. In our study, we utilized a comprehensive dataset of brain magnetic resonance imaging (MRI) scans to compare and assess the performance of different baseline AI models. These models included the K-Nearest Neighbors (KNN) Classifier, Logistic Regression, Decision Tree Classifier, and Multi-Layer Perceptron (MLP). Our analysis revealed that the KNN Classifier yielded the highest accuracy at 88.5%, making it the most suitable AI baseline model for brain tumor detection. These findings underscore the potential of AI models in achieving accurate and efficient brain tumor detection, paving the way for further advancements in this technology.",
      "project_yr": "2022",
      "project_quarter": "Winter",
      "graphic_link": "https://storage.googleapis.com/treatspace-prod-media/pracimg/u-2740/shutterstock_2140317261.jpeg",
      "headline": "Our study utilized a comprehensive dataset of brain magnetic resonance imaging (MRI) scans to compare and assess the performance of different baseline AI models.",
      "project_id": "b8f192e089d782f55aba98d180b19be10e08b51c",
      "expand": true,
      "tags": [
        "Neuroscience",
        "Biotechnology"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "publication_link": "https://www.jsr.org/hs/index.php/path/article/view/5601",
      "related_proj": [
        10,
        11,
        103
      ]
    },
    {
      "student_name": "Sam L.",
      "mentor_name": "Ivan Villa-Renteria",
      "mentor_title": "MS Computer Science at Stanford",
      "mentor_image": "assets/images/mentor_imgs/c3367c49c4a121b7eae2189efc7662edd8de4720.png",
      "student_image": "assets/images/student_imgs/c3e842b38827667b0ff8824ec9e23d2bcc1854ba.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Using Machine Learning to Detect Alzheimer’s Disease in MRI Scans",
      "project_desc": "Alzheimer’s Disease (AD) is a neurological disorder that slowly eats at the brain and affects a patient’s memory, thoughts, and behavior. It often develops in the later stages of life, and is a very heartbreaking disease to watch a loved one go through. As neurologists get closer to finding a cure for Alzheimer’s Disease (AD), it’s still necessary to catch the disease in its early stages to ensure the best quality of life for those who have it. Along with quality of life, it’s important to know if a patient has AD in order to protect them from autoimmune disorders that can worsen their symptoms or pose a threat due to their vulnerable state. We aimed to answer the question about if Magnetic Resonance Imaging (MRI) scans, which are often used in the diagnosing of other neurological disorders, can be used to diagnose AD in patients. While neurologists have already attempted this, we wondered if we could take it a step further by using machine learning to classify the data and separate it into different categories of dementia in order to properly diagnose a patient, as well as determine the severity of their AD. While conducting this research, we found surprising results, shown in how one of our inferior models reported nearly perfect accuracy while a model that was supposed to be superior reported a significantly poorer accuracy. Overall, our models reported back accuracies over 70%. These models also had to be incredibly complicated, with the image data being put into it having to be processed through hundreds of layers if we wanted a satisfactory accuracy. We concluded that while machine learning did report back very high accuracy, it is not a perfect tool for detecting fallacies in MRI scans, it is not meant to be the sole diagnostic for AD. ",
      "research_paper": "assets/pdfs/23349d22dd4128ee293754104a5f0dee4ea629b8.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "github": "https://github.com/SamLizotte/Machine-Learning-to-Detect-Alzheimers-Disease/tree/main",
      "graphic_link": "https://news.weill.cornell.edu/sites/default/files/styles/news_story_featured_image/public/story_featured_images/shutterstock_1111126898.jpg?itok=sURNjAQO&c=67d68c7ce0dba03a2ea246ef534ff826",
      "headline": "We aimed to answer the question about if Magnetic Resonance Imaging (MRI) scans, which are often used in the diagnosing of other neurological disorders, can be used to diagnose AD in patients. ",
      "project_id": "021f95a954214e2cd5744063595ab1f3bc876923",
      "expand": false,
      "tags": [
        "Neuroscience",
        "Diseases and Health Conditions"
      ],
      "published": true,
      "publication_link": "https://mareg5.zfairs.com/File/Client/efd07aa7-d1f7-43cc-8292-3bf60118f1f7?ext=.pdf",
      "related_proj": [
        10,
        137,
        133
      ]
    },
    {
      "student_name": "Aarav S.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Exploring Asteroid Orbits: Insights from Neural Network Modeling and Data-driven Analysis",
      "project_desc": "\tClassifying the orbits of asteroids contributes to research on the Solar System’s formation and evolution, making the classification of orbits a fundamental aspect of space exploration. However, accurate orbit determination is often hindered by skewed observational data stemming from uneven and limited sky coverage. In this study, orbital data from the NASA Jet Propulsion Lab was used in the classification of asteroid orbits through a machine-learning approach. Due to imbalances in the dataset, Synthetic Minority Oversampling Technique (SMOTE) was used to compensate for limited observations of certain orbit types. Afterwards, several classification models were implemented using specific orbital features. The most accurate results in this study were produced by a custom Keras neural network, with similar results demonstrated by an MLP neural network and k-nearest neighbors model. The custom neural network was able to effectively distinguish between orbital patterns, as demonstrated by its 99.18% validation accuracy and nearly diagonal confusion matrix. The capability of these models not only contributes to our understanding of asteroid orbits but also suggests potential improvements in orbit determination methodologies.",
      "research_paper": "assets/pdfs/4ca6f413ddb888b88f2bb34db01883b14b03534d.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://cdn.hswstatic.com/gif/gettyimages-1317904367.jpg",
      "headline": "In this study, orbital data from the NASA Jet Propulsion Lab was used in the classification of asteroid orbits through a machine-learning approach.",
      "project_id": "dbb1119fb778c16237dcc64e0d56987c55dc6f96",
      "expand": true,
      "tags": [
        "Astronomy",
        "Physics",
        "Science Fairs"
      ],
      "published": true,
      "science_fairs": "Accepted for Exhibition at TNJSF",
      "related_proj": [
        6,
        8,
        99
      ]
    },
    {
      "student_name": "Sam L.",
      "mentor_name": "Ivan Villa-Renteria",
      "mentor_title": "MS Computer Science at Stanford",
      "mentor_image": "assets/images/mentor_imgs/c3367c49c4a121b7eae2189efc7662edd8de4720.png",
      "student_image": "assets/images/student_imgs/c3e842b38827667b0ff8824ec9e23d2bcc1854ba.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Using Machine Learning to Detect Alzheimer’s Disease in MRI Scans",
      "project_desc": "Alzheimer’s Disease (AD) is a neurological disorder that slowly eats at the brain and affects a patient’s memory, thoughts, and behavior. It often develops in the later stages of life, and is a very heartbreaking disease to watch a loved one go through. As neurologists get closer to finding a cure for Alzheimer’s Disease (AD), it’s still necessary to catch the disease in its early stages to ensure the best quality of life for those who have it. Along with quality of life, it’s important to know if a patient has AD in order to protect them from autoimmune disorders that can worsen their symptoms or pose a threat due to their vulnerable state. We aimed to answer the question about if Magnetic Resonance Imaging (MRI) scans, which are often used in the diagnosing of other neurological disorders, can be used to diagnose AD in patients. While neurologists have already attempted this, we wondered if we could take it a step further by using machine learning to classify the data and separate it into different categories of dementia in order to properly diagnose a patient, as well as determine the severity of their AD. While conducting this research, we found surprising results, shown in how one of our inferior models reported nearly perfect accuracy while a model that was supposed to be superior reported a significantly poorer accuracy. Overall, our models reported back accuracies over 70%. These models also had to be incredibly complicated, with the image data being put into it having to be processed through hundreds of layers if we wanted a satisfactory accuracy. We concluded that while machine learning did report back very high accuracy, it is not a perfect tool for detecting fallacies in MRI scans, it is not meant to be the sole diagnostic for AD. ",
      "research_paper": "assets/pdfs/23349d22dd4128ee293754104a5f0dee4ea629b8.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "github": "https://github.com/SamLizotte/Machine-Learning-to-Detect-Alzheimers-Disease/tree/main",
      "graphic_link": "https://d2jx2rerrg6sh3.cloudfront.net/images/news/ImageForNews_768815_17049474459941307.jpg",
      "headline": "We aimed to answer the question about if Magnetic Resonance Imaging (MRI) scans, which are often used in the diagnosing of other neurological disorders, can be used to diagnose AD in patients.",
      "project_id": "021f95a954214e2cd5744063595ab1f3bc876923",
      "expand": false,
      "tags": [
        "Alzheimer's",
        "Diseases and Health Conditions",
        "Science Fairs"
      ],
      "published": true,
      "science_fairs": "2nd place at Massachusetts Region V Science Fair",
      "publication_link": "https://mareg5.zfairs.com/File/Client/efd07aa7-d1f7-43cc-8292-3bf60118f1f7?ext=.pdf",
      "related_proj": [
        68,
        4,
        178
      ]
    },
    {
      "student_name": "Emin C.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/fabc7768e36a22d4f25408e974243604e9913227.png",
      "domains": [
        "Business and Finance",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Artificial Intelligence in the Stock Market: Predicting Prices",
      "project_desc": "This research project focuses on stock price prediction through A.I. models as well as machine learning algorithms to maximize profit potential, improve investments,  and eliminate risk. Essentially, this project will demonstrate the modern implementation of A.I. in predicting the stock market. Potentially lucrative company stocks and shares have attracted investors as well as general interest in the stock market for decades, leading more people to try to predict the rise or fall of market prices. However, industry volatility and the seemingly unpredictable nature of the stock market have led many buyers to invest impulsively, sell their shares at the wrong time, or purchase stock from the wrong company. To combat these problems, we trained and tested A.I. models on our collected, classified data in order to generate accurate predictions. These models achieved average prediction errors of 0.12% for the stock prices of Amazon, 0.13% for the stock prices of Google, and 0.07% for Microsoft’s stock prices on the testing datasets.\n",
      "research_paper": "assets/pdfs/5f3f567abbe5616e25e2d29670bf1181e70f07d7.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://www.aljazeera.com/wp-content/uploads/2024/06/AFP__20240119__1945806302__v1__HighRes__StockIndexesRiseSP500ClosesAtTwoYearHigh-1718073351.jpg?resize=770%2C513&quality=80",
      "headline": "This research project focuses on stock price prediction through A.I. models as well as machine learning algorithms to maximize profit potential, improve investments,  and eliminate risk. ",
      "project_id": "ee8aff36bebcc433122a1e77eb970374a11c0aa1",
      "expand": false,
      "tags": [
        "Stocks",
        "Economics",
        "Published"
      ],
      "published": true,
      "publications": "The Stanford Journal of Science, Technology, and Society",
      "publication_link": "https://ojs.stanford.edu/ojs/index.php/intersect/article/view/3031",
      "related_proj": [
        2,
        8,
        14
      ]
    },
    {
      "student_name": "Seokhyun C.",
      "mentor_name": "Victoria Lloyd",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/a05002f29eb54c606f45e6ae8d8d7d6b9885493b.png",
      "domains": [
        "Environment",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Predicting Climate Change Using an Autoregressive Long Short-Term Memory Model",
      "project_desc": "Climate change is an issue that has and will affect humanity throughout the time that we are on Earth. Due to this fact, scientists have created numerous methods of modelling the climate, with mathematical models being used most widely. However, due to some limitations that these models hold, researchers have turned to utilize machine learning models to predict the future climate. This is because AI has a reputation for being able to handle complex data and recognize patterns that humans previously could not have. This study aims to create a baseline machine learning model that utilizes an Autoregressive Recurrent Neural network with a Long Short term memory implementation for the purpose of predicting climate. This type of model architecture has been utilized for other applications that relate to time series data, yet it has yet to be attempted to be used in the context of climate predictions. The data that was utilized was retrieved from the ensemble mean version of the ERA5 dataset. The model created from this study was able to predict the general trends of Earth(i.e, the poles are cold, and the equator region is warm) for both when predicting the climate and when it was predicting weather. When predicting the climate, the model was able to have fair accuracy for a long period of time, with the ability to predict seasonal patterns. This feat is one that other researchers were not able to do with the complex reanalysis data that this study has utilized. This work demonstrated that this type of model can be utilized in a climate forecasting approach as a viable alternative to mathematical models and can be utilized to supplement current work that is mostly successful in short term predictions. ",
      "research_paper": "assets/pdfs/00c71309483550d6ac87c750f83854524d73733b.pdf",
      "project_yr": "2022",
      "project_quarter": "Winter",
      "graphic_link": "https://ca-times.brightspotcdn.com/dims4/default/06cdef8/2147483647/strip/true/crop/2696x1664+0+0/resize/1200x741!/quality/75/?url=https%3A%2F%2Fcalifornia-times-brightspot.s3.amazonaws.com%2Fa4%2Faf%2F13fe5626461b9e801ac6e5161c51%2Funited-states-arctic-62857.jpg",
      "headline": "This study aims to create a baseline machine learning model that utilizes an Autoregressive Recurrent Neural network with a Long Short term memory implementation for the purpose of predicting climate.",
      "project_id": "598032afba3b6220dcc69ec7f0c0fd101475e823",
      "expand": true,
      "tags": [
        "Climate Change",
        "Sustainability",
        "Published"
      ],
      "published": true,
      "publications": "Frontiers in Environmental Science",
      "publication_link": "https://www.frontiersin.org/articles/10.3389/fenvs.2024.1301343/full",
      "related_proj": [
        113,
        95,
        70
      ]
    },
    {
      "student_name": "Arnav J.",
      "mentor_name": "Erick Siavichay",
      "mentor_title": "Computer Science Alum from Stanford, Software Engineering and Data Science at Tagg",
      "mentor_image": "assets/images/mentor_imgs/788ef03445f644b0e53860927d1ce0d68eeac867.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "Evaluating the Efficacy of the 3D U-Net Architecture For Glioblastoma Multiforme Tumor Segmentation",
      "project_desc": "Glioblastoma is the deadliest form of brain cancer which begins as a congregation of cancerous cells within the brain but then progresses into invading and destroying healthy brain tissue [9]. Radiation therapy, the most popular treatment option, is where neuro-oncologists apply intense radiation energy beams directly on the tumor region to kill the cancerous cells. However, for radiation therapy to be effective, the segmented magnetic resonance image (MRI) in which the oncologists base where to apply the radiation must be segmented nearly 100% accurately, or else the energy beams will mistakenly damage healthy brain tissue. Additionally, radiation therapy will be futile if the MRI segmentation is not complete in time for the therapy appointment. Our research problem is evaluating the performance of notable segmentation models for accurate and immediate 3D segmentation of glioblastoma multiforme brain tumors from MRIs while focusing on one particular architecture, the 3D U-Net. Each of the models was given 850 MRIs from the BraTS 2021 dataset, which is an annual competition hosted by the University of Pennsylvania [8]. Our 3D U-Net model achieved a testing accuracy of 0.986) accuracy and took approximately 55 seconds to predict segmentations of the tumor region. The results reveal that the 3D U-Net model is capable of automating glioblastoma tumor segmentation in significantly fewer hours than a human oncologist would take, all while maintaining similar or higher accuracy, where even a minor difference can mean the distinction between life and death.",
      "research_paper": "assets/pdfs/97798e5c4041b6014d8653357f66d617d3f55498.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://www.genengnews.com/wp-content/uploads/2023/09/GettyImages_1190794650_BrainTumor_cropped.jpg",
      "headline": "This research evaluates the performance of the 3D U-Net model for automated glioblastoma tumor segmentation from MRIs, achieving 98.6% accuracy and significantly faster processing times than human oncologists, crucial for effective radiation therapy.",
      "project_id": "7197e6be056bf44a07ddae9b38e3c556c4d93ec9",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "published": true,
      "publications": "Journal of Student Research",
      "publication_link": "https://www.jsr.org/hs/index.php/path/article/view/6558",
      "related_proj": [
        164,
        125,
        32
      ]
    },
    {
      "student_name": "Suhaan K.",
      "mentor_name": "Tony Rodriguez",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/b985b04a876866abc4fd7da68d847f15e821bbe2.png",
      "domains": [
        "Physics",
        "Published Papers and Science Fairs"
      ],
      "project_title": "The Use of Artificial Intelligence in Gravitational Microlensing Detection for Dark Matter Discoveries ",
      "project_desc": "Gravitational lensing has been used for over three decades as effectively the only tool to reveal the nature of \"dark\" sources in the Milky Way. \"Dark\" sources include stars, planets, black holes, or even dark matter. We hypothesized that by using 600,000 stars as data obtained from The Zwicky Transient Facility, a telescope at the California Institute of Technology, we could use Artificial Intelligence to increase the speed and accuracy of microlensing detection. The main objective of this project was to distinguish between microlensing and non-microlensing events, the former of which could be dark matter events. Through the rigorous testing of various Artificial Intelligence models, including Linear Regression, KNN, SVC, Decision Tree, and others, we conducted a comparative analysis to discern disparities in accuracy, precision, and other recall among these models Through hyperparameter tuning and the elimination of false positives, the KNN model was confirmed to be the most accurate model, generating a 97% accuracy for the detection of microlensing events within the Milky Way. This scientific inquiry aims to improve future searches for microlensing events, ultimately expediting the current scientific processes employed in microlensing detection and helping astronomers place search constraints on the nature of dark matter through AI.",
      "research_paper": "assets/pdfs/b2fb21db6acbc409292115f0d6f07206b1761e79.pdf",
      "project_yr": "2025",
      "project_quarter": "Spring",
      "graphic_link": "https://i.natgeofe.com/n/f9954a0f-f88e-4ebe-b2c0-39d531a95afc/48384_16x9.jpg",
      "headline": "This research utilizes Artificial Intelligence to enhance the speed and accuracy of microlensing detection using data from 600,000 stars observed by The Zwicky Transient Facility. Through comparative analysis and hyperparameter tuning, the KNN model was identified as the most accurate, achieving 97% accuracy in detecting microlensing events, which may include dark matter, within the Milky Way.",
      "project_id": "ce2b76a0ff00e0d1bd386c2afaebfb5f77484b4d",
      "expand": true,
      "tags": [
        "Engineering",
        "Astronomy",
        "Physics"
      ],
      "published": true,
      "science_fairs": "3rd Place in the Texas Science and Engineering Fair",
      "publication_link": "https://txsef.tamu.edu/awards/2024-category-awards/",
      "related_proj": [
        16,
        12,
        36
      ]
    },
    {
      "student_name": "Azaan R.",
      "mentor_name": "Joe Xiao",
      "mentor_title": "PhD, Data Scientist at Optum, prior lead AI/Data Scientist at Target and General Mills",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Published Papers and Science Fairs"
      ],
      "project_title": "VisionAssist: Enhancing Accessibility for Individuals with Visual Impairment Through AI",
      "project_desc": "Millions of people around the world are partially or totally blind. Such people face significant challenges in their educational pursuits and therefore they often struggle in life. Can Artificial Intelligence and latest developments in technology help visually impaired people understand literary content easier? To address this question, I have developed an Artificial Intelligence based proof of concept system that accurately converts any image with textual/mathematical data into Braille and Audio mediums as close to real-time as possible. The system takes an image as input and extracts text from it through optical character recognition (OCR) using a finetuned and optimized version of opensource PaddlePaddle OCR model. The system then has two different output options, an audio-based format, and a braille-based tactile format. Word Error Rate (WER) and Character Error Rate (CER) were used as evaluation metrics to evaluate performance of the system on different types of data.  My system achieves a weighted average WER of approximately 7% with an average processing time of around 1 sec per 500 characters. The evaluation data shows that the system achieved the goal of having an accurate detection of the data in near real-time speed. My prototype outperforms much bigger and commercial models like ChatGPTv4 in terms of latency as well. This project shows that technology and AI can be an additional tool to help visually impaired people better understand educational content which can eventually uplift them in life.",
      "project_yr": "2025",
      "project_quarter": "Spring",
      "graphic_link": "https://lhblind.org/wp-content/uploads/2023/10/AI-AdobeStock_567681994-web.jpg",
      "headline": "This project explores how AI can support individuals with visual impairments by developing a system that converts images containing text or math into audio and Braille in near real-time. Using a fine-tuned OCR model, the system achieves high accuracy and low latency, demonstrating that AI can be a powerful tool for improving accessibility to educational content.",
      "project_id": "4064cb21920d9ef34392a083b12c6a426307a1e4",
      "expand": true,
      "tags": [
        "Eyes",
        "Diseases and Health Conditions",
        "Engineering"
      ],
      "published": true,
      "science_fairs": "1st Place in the Alameda County Science Fair",
      "publication_link": "https://docs.google.com/presentation/d/1A3VusY0XctXC1fBKNjxb-26fjAWup7ajJel2ZtFxVsY/edit?slide=id.g335de835b50_0_965#slide=id.g335de835b50_0_965",
      "related_proj": [
        97,
        135,
        67
      ]
    },
    {
      "student_name": "Anika S.",
      "mentor_name": "Ying Hang Seah",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/3e81a81724cf0073d0800f5d96bcfd5615110f17.png",
      "domains": [
        "Environment",
        "Engineering"
      ],
      "project_title": "Landfill Net: A Convolutional Neural Network (CNN) Architecture for the Detection of Landfills from Satellite Imagery in the Continental United States",
      "project_desc": "As we continue through the 21st century, the effects of climate change have become more and more apparent, with methane emissions playing a pivotal role as one of the most potent greenhouse gasses. Landfill decomposition is a large contributor to this rise in methane, and certain landfills have even adopted methane capture systems that enable the conversion of this methane into valuable energy resources. To continue to mitigate these emissions, it is imperative to establish a comprehensive United States landfill database. We present a Convolutional Neural Network (CNN) model that uses image data in the United States to distinguish whether an image is a landfill or not. The model had a test accuracy of 97.1%, a precision of 96.3%, a recall of 98.1%, and an F-1 score of 97.2%. The results show that the model holds the potential to uncover illegal landfills within the United States and facilitate an informed expansion of methane collection systems in landfill sites.",
      "research_paper": "assets/pdfs/9b7a9bb8993cb8645f1200f22b54f5e7d19beefc.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "github": "https://github.com/anikaSeshan/LandfillNetResearch.git",
      "graphic_link": "https://img.freepik.com/free-vector/car-dump-cartoon-composition-with-crushed-auto-landfill-vector-illustration_1284-81099.jpg?semt=ais_hybrid&w=740",
      "headline": "This project presents a Convolutional Neural Network (CNN) model that identifies landfills in the U.S. using image data, achieving a 97.1% test accuracy. By supporting the creation of a national landfill database, the model can help detect illegal sites and guide the expansion of methane capture systems to reduce greenhouse gas emissions.",
      "project_id": "cb6efe28ac166c1553671ee2e20ce03f130ee44f",
      "expand": true,
      "tags": [
        "Sustainability",
        "Pollution",
        "Engineering"
      ],
      "published": true,
      "science_fairs": "Tackling Social Issues with AI | Inspirit AI Research Symposium | Category Winner",
      "related_proj": [
        70,
        159,
        125
      ]
    },
    {
      "student_name": "Arpan A.",
      "mentor_name": "Mirna Kheir Gouda",
      "mentor_title": "Biomedical Engineering PhD Candidate at MIT",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/649fdf32d8c6917b69fb059ba2d77c623865341c.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Most Important Soil Properties to Consider for High Crop Yield",
      "project_desc": "With the current population of Earth massively growing, there is a real risk coming up in the near future of not having enough food to feed the planet. One of the best ways to solve this upcoming problem is to improve the quality of the soil because that will help increase Earth’s overall crop yield in a sustainable way. However, figuring out how to improve soil can be a big challenge since there are so many elements and properties of soil to analyze. To help solve this problem we looked into the elements or properties of soil that could have the biggest influence on soils ability to be productive for growing crops. We found online datasets and used unsupervised machine learning by applying both principal component analysis (PCA) and hierarchal clustering on our data. We concluded that calcium, potassium, pH, and soil organic carbon are the elements or properties of soil that have the biggest influence on soil's ability to be productive for growing crops. If these results are accurate, it can massively help many farmers and people in improving their soil quality. This can lead to us taking a big step in the direction of increasing Earth’s overall crop yield.  ",
      "research_paper": "assets/pdfs/6db01dae04a368907c9e65aed25f76c0199aa535.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.allaboutarizonanews.com/wp-content/uploads/2019/01/CottonField.PhoenixAZ.140320-1-1080x608.jpg",
      "headline": "With the current population of Earth massively growing, there is a real risk coming up in the near future of not having enough food to feed the planet. One of the best ways to solve this upcoming problem is to improve the quality of the soil because that will help increase Earth’s overall crop yield in a sustainable way.",
      "project_id": "c8ddc52adbc15aa68fc701cde83fd2e65d2e5fcd",
      "expand": false,
      "tags": [
        "Earth Systems",
        "Irrigation",
        "Water Use"
      ],
      "related_proj": [
        63,
        123,
        33
      ]
    },
    {
      "student_name": "Leah N.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/23725615a8586e726fc6fa97c0ccdc962d6825bd.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Developing an Accurate AI Algorithm for Histopathologic Cancer Detection",
      "project_desc": "\tThis paper discusses the development of a machine learning algorithm that accurately detects metastatic breast cancer (the cancer has spread elsewhere from its origin part) in select images that come from pathology scans of lymph node sections. Being able to develop an accurate artificial intelligence (AI) algorithm would help significantly in breast cancer diagnosis since manual examination of lymph node scans is both tedious and oftentimes highly subjective. The usage of AI in the diagnosis process provides a much more straightforward, reliable, and efficient method for medical professionals and would enable faster diagnosis and, therefore, more immediate treatment. The overall approach used was to train a convolution neural network (CNN) based on a set of pathology scan data and using the trained model to binarily classify if a new scan were benign or malignant, outputting a 0 or a 1, respectively. The final model’s prediction accuracy is very high, with 100% for the train set and over 70% for the test set. Being able to have such high accuracy using an AI model is monumental in regards to medical pathology and cancer detection. Having AI as a new tool capable of quick detection will significantly help medical professionals and patients suffering from cancer.",
      "research_paper": "assets/pdfs/8a5e8935253f9eab111645e70bd509856ac1b1f8.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://miro.medium.com/max/800/1*0YHaGNOMIhZaHkZ7RR-FFQ.jpeg",
      "headline": "In this specific research project, we will be focusing on the lymph node scans of women with breast cancer, which is the most common cancer for women residing in the US, other than skin cancer. Research statistics show that about 1 in 8 women in the United States will develop invasive breast cancer throughout her life.",
      "project_id": "497d2dfb6fdc004c66430347d7e0e2026a04d6c5",
      "expand": false,
      "tags": [
        "Cancer",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        163,
        156,
        7
      ]
    },
    {
      "student_name": "Jason M.",
      "mentor_name": "Anna Orosz",
      "mentor_title": "Data Science MS from UPenn, Machine Learning Engineer at LinkedIn",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/938de8743e5016120fb1d9f657fbd8f531a0858d.png",
      "domains": [
        "Sports and Motion Sciences",
        "Engineering"
      ],
      "project_title": "Allez Go: AI Fencing Referee",
      "project_desc": "The gradual increase in online fencing videos over the past decade has allowed for novel technical projects in fencing that rely heavily on data, such as artificial intelligence.  This study resulted in a state-of-the-art lightweight Temporal Convolutional Network to referee fencing bouts and classify actions as either a touch for the fencer on the left or the fencer on the right. To address this problem, we developed a pose estimation and audio analysis approach to autonomously referee fencing bouts. Using a custom dataset of international level fencing from the last 7 years, including ~4000 unique clips, our model achieved an accuracy of 89.1%, a 20% increase over previous state-of-the-art models. This model leverages advancements in human pose estimation to extract the position of both fencers and avoids high computational loads typically associated with CNNs. Additionally, it uses a novel technique to solve the issue of blade contact, a key component of refereeing fencing that was generally unaddressed in previous works. Our novel solution uses audio to ‘listen’ for the sound of blade contact rather than attempting to identify it visually. ",
      "research_paper": "assets/pdfs/06bec52004cebec38e16889039e8afd90442d3c9.pdf",
      "project_yr": "2022",
      "project_quarter": "Spring",
      "project_webpage": "https://bananapeel202-fencingwebapp-upload-xnj04o.streamlitapp.com/",
      "graphic_link": "https://images.theconversation.com/files/133347/original/image-20160808-18050-xlihp1.jpg?ixlib=rb-1.1.0&q=45&auto=format&w=1200&h=900.0&fit=crop",
      "headline": "Technology in fencing is generally an underdeveloped field and automated referees present potentially significant benefits to the sport. Automated referees will offer a more consistent call compared to a group of human referees with slightly different interpretations of the fencing rules.",
      "project_id": "77ececfe30468cb64dc87fadf0e983552743a873",
      "expand": false,
      "tags": [
        "Sports",
        "Fencing",
        "App Development",
        "Audio Analysis"
      ],
      "related_proj": [
        49,
        141,
        167
      ]
    },
    {
      "student_name": "Mike S.",
      "mentor_name": "Matthew Radzihovsky",
      "mentor_title": "Computer Science MS from Stanford, AI/ML Engineer at Apple",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Using Machine Learning Architecture to Detect Brain Tumors",
      "project_desc": "Magnetic Resonance Imaging (MRI) scans are the primary way medical professionals study the brain. Doctors are trained to identify irregular cell growth patterns and notify patients if they have or are at risk of cancer. Identifying tumors before they become malignant saves both resources and welfare of patients. In this paper, we apply a basic SKLearn Logistic Regression model, a custom Convolutional Neural Network model, and a transfer learning model based on the VGG16 model to automatically diagnose malignant or benign tumors in MRI scans. We found that the VGG16 model performed the best, with a validation accuracy of 91.6%. Logistic Regression had a validation accuracy of 82.3% and CNN an accuracy of 83.2%. These promising results suggest that automated machine learning models can be used as a MRI scan analyzer to diagnose patients more often and more quickly which may have the ability to identify tumors before they become malignant, saving lives.",
      "research_paper": "assets/pdfs/9b4c949205be279ffa61b4678e8e0dacebe2e815.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "http://www.roswellpark.org/sites/default/files/shutterstock_44859535.jpg",
      "headline": "Not all tumors in the brain are cancerous, so irregular cell growths can often be determined as benign and are ignored completely. However, many of these low-danger cell growths can develop into higher-danger cancerous tumors. Identifying these low-danger tumors early is key to preventing cancer in the future.",
      "project_id": "358df9da416a0cea3a9ae9e2232c1b6fd978dde5",
      "expand": false,
      "tags": [
        "Cancer",
        "Brain Cancer",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        42,
        134,
        183
      ]
    },
    {
      "student_name": "Fiona B.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Using Machine Learning Models to Determine Medical Features Most Indicative of Gestational Diabetes",
      "project_desc": "Since not all females in America have access to abortions, my research is important because it can lead to ways to minimize dangerous and potentially fatal risks in pregnancies. Gestational diabetes is a type of diabetes that develops during pregnancy that can lead to many health concerns for both the mother and fetus. Machine learning classification can take many different features or inputs to guess what the output should be, so in the case of gestational diabetes, in a ML were given many medical features and had to predict what the outcome could be, someone could figure out what the ML thought was most important, and that could be an indicator of what medical factors are closely connected with a disease. I used four different machine learning models to each predict if a patient given a list of their medical features had gestational diabetes. I then found out which features the models relied on to make their guess in order to determine the most indicative medical factors. All models agreed that glucose was a medical factor that they relied on to make their predictions. Two different models had the exact same three features: glucose, BMI, and age. In conclusion, if medical professionals know that these three factors are most indicative of someone developing gestational diabetes, then these areas can be targeted when someone is trying to get pregnant to minimize the likelihood of developing gestational diabetes.",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://assets.medpagetoday.net/media/images/100xxx/100274.jpg",
      "headline": "With the recent overruling of Roe v. Wade, pregnancies have become especially dangerous since many females who previously abort fetuses no longer can. By doing this research project, I can help figure out what medical factors may lead to gestational diabetes, and hopefully from there, the results can be used to counteract those features.",
      "project_id": "5e84d8f6b96a6f4606997b2bff26e36a6635f559",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Diabetes"
      ],
      "related_proj": [
        98,
        31,
        163
      ]
    },
    {
      "student_name": "Audrey W.",
      "mentor_name": "Matthew Radzihovsky",
      "mentor_title": "Computer Science MS from Stanford, AI/ML Engineer at Apple",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Prediction of Nitrogen Dioxide Level using Machine Learning Models",
      "project_desc": "Air pollution has been a lingering problem to many developing countries. Overpopulated cities and unregulated industrial and vehicle emissions constantly destroy the natural environment and harm human health. Being one of the most common air pollutants, nitrogen dioxide, also known as NO2, damages our respiratory system, destroys aquatic food chains, and contributes to the formation of acid rains. In this paper, linear regression, random forest regressor and decision tree regressor from scikit learn are used to predict NO2 level in parts per million (ppm) by analyzing its correlation with other air pollutants in a specific city. The model achieved an accuracy of 69,1 percent using  random forest and a 66 percent accuracy using decision tree. The accuracy of the model can increase with a more detailed dataset, but further scientific research and discovery are required to actually predict acid rain based on just NO2, since the amount of NO2 that will cause acid rain is uncountable and many more variables go into the formation of acid rain. ",
      "research_paper": "assets/pdfs/f6fb83c933cc866c2ba1130446fe194173bd6b18.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://images.newscientist.com/wp-content/uploads/2017/05/02163148/p8713477-2.jpg",
      "headline": "As we release more pollution into the atmosphere as factory and vehicle byproducts, acid rains are formed and are constantly damaging the environment and harming wildlives under and above the ocean. If the occurrence of acid rains can't be limited, the phenomenon of this man-made disaster will threaten the health and safety of billions. By predicting the potential development of acid rain  with nitrogen dioxide (NO2) levels, one of the major components of acid rain, we could prevent it by limiting the number of specific gasses produced.",
      "project_id": "06ce4752b575b0bee26c2b7e67d42f6b372107ab",
      "expand": false,
      "tags": [
        "Earth Systems",
        "GHGs",
        "Pollution"
      ],
      "related_proj": [
        120,
        67,
        54
      ]
    },
    {
      "student_name": "Keithan P.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/student_imgs/9802270da4fc207a5bb544c48b138c699262efcb.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Predicting Skin Cancer using Machine Learning",
      "project_desc": "The motivation for this research project was to create an accurate model that could be used to identify the presence of different types of skin cancers. The latter objective is to create an application that could be easily used in rural areas, as they do not have access to modern medical technology. The approach began with loading the dataset into a readable format. I used the load_dataset import from Hugging Face [4] to complete this task. The models that were used for this research project were all from Hugging Face [4]. The highest accuracy was 0.8633 and the lowest accuracy was 0.6647. The models were all run on a colab notebook. The meaning of these results is that on the best model the image was predicted correctly 86% of the time. The conclusion of this research is that more research must be conducted and more models must be used in order to achieve an accuracy that would warrant the model to be used in the real world.",
      "research_paper": "assets/pdfs/7fc8bcd4e544090345a2eada393c88b1c67aed1c.pdf",
      "project_yr": "2022",
      "project_quarter": "Spring",
      "graphic_link": "https://images.everydayhealth.com/images/how-is-skin-cancer-diagnosed-722x406.jpg",
      "headline": "Skin cancer affects an increasing number of people which is the main motivation for my research. The goal is to develop a model that is accurate enough for real world application and that could be distributed to more rural areas which have decreased access to medical technology.",
      "project_id": "949e6f3e9e51ef95bc5426b1288aa28d9cd465d9",
      "expand": false,
      "tags": [
        "Skin Cancer",
        "Diseases and Health Conditions",
        "Cancer"
      ],
      "related_proj": [
        24,
        167,
        44
      ]
    },
    {
      "student_name": "Sohum T.",
      "mentor_name": "Akshay Jagadeesh",
      "mentor_title": "Computational Neuroscience PhD from Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/1cf575a6af76512e25f671e8f8a2e2ebf39f9af9.png",
      "domains": [
        "Psychology",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "How Does One’s Background Determine Their Mental State?",
      "project_desc": "Is there a correlation between one’s mental health status and one’s Age, Family History, Workplace Environment, or other aspects about one’s life? This paper answers this question by using responses to a survey that over 1000 people took. The participants of this survey were of diverse backgrounds and had varying answers. Models such as Logistic Regression, the Decision Tree Classifier, and the SVC model were used to produce correlations. The results indicate that there is a positive correlation between one’s family history of mental health and one’s current mental state, and that there is a negative correlation between one’s work interference and one’s current mental state, among others.",
      "research_paper": "assets/pdfs/44369ac3047fac0f55d7d730ab7b8c465404cb6a.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://hbr.org/resources/images/article_assets/2018/10/Oct22_18_862457080.png",
      "headline": "Mental health issues have become very prevalent in recent times, and although significant progress has been made in terms of treatment in the form of counseling, medication, and other methods, in order to truly find out the root cause of many mental health issues, a correlation has to be drawn between one’s mental state and another factor, such as family history, age, work environment, etc. It is also beneficial to correlate one’s mental state with a multitude of factors to see the compilation.",
      "project_id": "6021eb4f2c7e1f36f6d75c3ba4ea1a05c93bcc50",
      "expand": false,
      "tags": [
        "Mental Health",
        "Psych"
      ],
      "related_proj": [
        84,
        52,
        126
      ]
    },
    {
      "student_name": "Eesha S.",
      "mentor_name": "Victoria Lloyd",
      "mentor_title": "PhD Student in Physics, Prior Research Collaborator at Caltech, ",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Gamma Ray Classification",
      "project_desc": "The focus of this research project is to be able to classify high-energy gamma particles as signal or background. It is important to solve this issue, as it allows us to explore phenomenas that we are unable to see on Earth. High energy gamma particles are also a major part of particle physics. Research in particle physics has helped us in the world of medicine, in drug development, along with exploring more about matter that makes up our world. I created a model that classifies a particle as signal or background based on given test features that I was able to get from the NASA MAGIC Gamma Telescope experiment. This was accomplished by creating an ML model and following the process of training, predicting, and evaluating to test various models. I found the accuracy score of the model, created a confusion matrix, and calculated the f1 score for all the models. For the final model, I got an accuracy score of 88.44% and an f1 score of about 0.883, allowing me to finally find a model that classifies the high energy gamma particles with the highest accuracy. ",
      "research_paper": "assets/pdfs/1b98ab588e67f604859ebb273dcb5e3e5affc1c8.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://scx2.b-cdn.net/gfx/news/2020/aregammarayb.jpg",
      "headline": "This research paper is in this domain of physics, specifically using Artificial intelligence (AI) to classify high-energy Gamma particles as background or signal. I was able to create an AI model to classify these gamma particles with high accuracy. I was able to accomplish this by using a wide range of data, from the MAGIC Gamma Telescope Dataset, which consists of key features, to construct a model.",
      "project_id": "22b10a274fc2ce6a3c295b663210ee33b1ec2c1f",
      "expand": false,
      "tags": [
        "Astronomy",
        "Gamma Rays"
      ],
      "related_proj": [
        151,
        150,
        166
      ]
    },
    {
      "student_name": "Molly H.",
      "mentor_name": "Sharon Chen",
      "mentor_title": "Electrical Engineering and Neuroscience MS from Caltech",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Identifying Parameters in Water Potability Analysis Through Machine Learning",
      "project_desc": "Machine learning has become a rapid and prevalent tool in the growth of environmental science. With the onset of emerging data, data analysts have relied on ML models to reveal hidden trends. In past papers, machine learning has been brought up countless times as an innovative method to rapidly sort through data in big groups. However, developing an artificial neural network model is a sparsely used tool in research due to its lack of human interface, risk factors, and not enough experimentation in real-life applications. In this research, we implemented multiple ANN models to identify factors in a water’s chemical levels to identify which has the largest weight or bias in determining whether the water is potable or not. Subsequently, we created a model that uses these parameters to analyze and predict whether a water source will be safe to drink. Water is an essential resource for all beings on Earth, and as the climate changes, the water levels rise, and along comes pollutants to once safe drinking water. By comparing the accuracy between both classifier and regressor ML models, each model was hyper-parameterized to develop a better accuracy, as well as identify the weight each parameter carries towards the potability of the source. The results demonstrate a promising conclusion that the number of solids affects the potability versus other chemical levels. That said, there was not a strong, distinct correlation and every factor plays an important part. The accuracy of these models returned a sub-par 64.77% accuracy with a 0.53 precision, a 0.46 recall score, and a 0.49 f1 score. In conclusion, while the accuracy of the models is ineffective to be used just yet on a bigger scale, they offer a stepping stone in the development of machine learning models in the environmental science fields, resourcing digital infrastructures to analyze and dissect the Earth and its trajectory through climate changes.",
      "research_paper": "assets/pdfs/351f19c3c6efe36fbd516eddb955f804b55c3aaa.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.prominent.com/media/References/potable-water-stadtwerke-arnsberg_Header_1.jpg",
      "headline": "Predicting whether the water is potable or not can be helpful for people who are reliant on bodies of water and redirect them to safer options. It will also be beneficial to apply the algorithm to other places where it is expensive and inefficient to send people out and collect water samples. Over the past couple of decades, researchers have often commented on the lack of funding as a source of error when it comes to data analysis and the accuracy of the research.",
      "project_id": "6fa1b51e4ed9c740192fc78b3337892731c0a76a",
      "expand": true,
      "tags": [
        "Earth Systems",
        "Water Use"
      ],
      "related_proj": [
        172,
        106,
        9
      ]
    },
    {
      "student_name": "Anitej S.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/student_imgs/6ef1f4892c6d91e8d1c41242eda56a902f2c3807.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": " Is GPT-3 smarter than a sixth-grader?",
      "project_desc": "I work with the Davinci model in GPT-3, an autoregressive language model, to answer middle-school science textbook questions in the textbook question answering (TQA) dataset. The dataset was split into training, test, and validation sets. In this task, I simulate a student taking a test in three different scenarios. First, the Zero-Shot-Learning experiment where I only provide the model with the questions from a specific lesson. This would be the equivalent of a student going into a test without studying as the model has not gathered any knowledge from the lesson. Second, the Few-Shot-Learning experiment, where I provide the model with specific lesson content from the textbook and the corresponding questions. This equates to a student skimming over the lesson content before taking the test. Lastly, I fine-tuned the Davinci model on some of the textbook questions and then fed it questions. This is similar to a student doing a thorough review of the material before taking the test. After conducting all three experiments, I compare their accuracies and in doing so, highlight the “intelligence” and limitations of GPT-3. \n",
      "research_paper": "assets/pdfs/bb454b71ae4f6b392600ee1ea2ba28d5b74d4841.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/TejSuklikar/GPT-3-Research-Project",
      "graphic_link": "https://decemberlabs.com/wp-content/uploads/2021/02/768_2zpph46r1elg.jpg",
      "headline": "Question answering (QA) and Large Language models (LLM) have been a major research focus in Artificial Intelligence for several years. In 2017, a task called Textbook Question Answering (TQA) was introduced. The task included lessons from a middle school science textbook consisting of texts, diagrams, and natural questions. Many people attempted to create question answer models but reported sub-par accuracies.",
      "project_id": "12bafc1b0d40c5efca04d8d8aaef32d3ba22e042",
      "expand": false,
      "tags": [
        "GPT-3",
        "Natural Language Processing"
      ],
      "related_proj": [
        118,
        87,
        66
      ]
    },
    {
      "student_name": "Adam S.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/530e0699a5b243cc889e69d330ab5532879e7043.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Identifying Cancer Types in Microscope Images of Lung and Colon Cells",
      "project_desc": "This program seeks to reach a high accuracy of detecting types of cancer in addition to whether or not cells are benign. The model categorizes the cell images into 5 classes: benign colon, benign lung, cancerous lung adenocarcinoma, cancerous colon adenocarcinoma, and cancerous lung squamous cell carcinoma. Two models are then trained with the images: a random forest classifier, and a deep neural network using convolutional layers. Finally a third model predicts which classifier is more likely to make the correct prediction for a given image, and that classifier’s prediction is used. This tactic allows the models to learn different patterns, but still be effective at predicting all classes of images.\n",
      "research_paper": "assets/pdfs/4f7ae05d6bc184d7be2a9bed0b3fb13d758df0fe.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.pulmonologyadvisor.com/wp-content/uploads/sites/21/2019/02/chestxrayscang869879614_1362374.jpg",
      "headline": "Lung adenocarcinoma cells make up large proportions of lung cancer cases and colon cancer is one of the most prevalent cancers in the United States. Although many different medicines have been recently developed to attack these cancers, the most effective way to stop it is early detection.",
      "project_id": "ead7c7979987656fc00f4364ca8cbc9d2331b23e",
      "expand": false,
      "tags": [
        "Cancer",
        "Diseases and Health Conditions",
        "Lung Cancer"
      ],
      "related_proj": [
        19,
        10,
        101
      ]
    },
    {
      "student_name": "Bertran M.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Psychology",
        "Education"
      ],
      "project_title": "Detecting Facial Expressions",
      "project_desc": "The research question put forward by this paper is “To what extent can emotion detection artificial intelligence algorithms be used in order to improve education, business success, employee safety, healthcare and video game testing?” This paper looks to be able to give the consumers of the solution the ability to pinpoint which emotions educators, healthcare workers, video game testers and consumers of their business are feeling in order to improve their quality. In all of the areas that have been previously mentioned, there is a large inability to be able to identify when certain workers are feeling certain emotions, as a result of lack of recall memory, lack of honesty, or simply lack of awareness. In order to identify the satisfaction and dissatisfaction of healthcare workers, educators, video game testers and consumers of a business at certain moments, the quality of healthcare, education, video games and business websites will all increase respectively by being able to identify when workers are feeling a certain way, and how to reinforce or reduce the emotions being felt during those moments. \nIn order to solve this problem, many classification algorithms will be explored, such as convolutional neural networks, K nearest neighbor, multi-layer perceptron classifier, amongst many others. These classification algorithms will look for a successful and quality classification of the images being presented in order to ensure excellence whenever this solution is implemented in the areas mentioned previously. \n\tBy using the algorithms mentioned previously, the accuracy of the model was significantly higher than what it would have been had it been chosen randomly. For instance, with a convolutional neural network, there was an accuracy of 25%, whereas random assignment would have had a success rate of 14.29%\nThe conclusions from this investigation are that artificial intelligence algorithms can be used in order to classify images into 7 categories according to the emotion being displayed on them, and this solution can and will be successfully implemented into the areas mentioned previously. \n",
      "research_paper": "assets/pdfs/aa2e01c84ca3e24e1bb376ba5988cfc10d831d4d.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://post.psychcentral.com/wp-content/uploads/sites/4/2021/07/young-man-portrait-looking-at-camera-blank-expression-1200x628-facebook-1200x628.jpg",
      "headline": "This paper shows that machine learning algorithms can be used in order to classify emotions into different categories. The most promising results were from the K-nearest neighbor model, with 61.5% accuracy in the training set, and 32.1% accuracy in the testing set.",
      "project_id": "eb645ba3e84ff785a70b7e6d92d0b9aacab60af8",
      "expand": false,
      "tags": [
        "Psych"
      ],
      "related_proj": [
        86,
        58,
        61
      ]
    },
    {
      "student_name": "Ian C.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/bce6dade3cedd6222d229fb853ade51f61262b17.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Using Logistic Regression in the Early Detection of Tsunamis caused by Earthquakes",
      "project_desc": "Tsunamis have forced large financial investments devoted to their preparation and repair. One primary cause of such tsunamis are earthquakes, often those occurring nearby. By using processed NCEI/WDS datasets for earthquakes and tsunamis, we built a logistic regression model to predict the probability a tsunami would form following a given earthquake using features such as magnitude and location with an accuracy of around 80%. In addition, a linear regression model was used to predict the resulting tsunami’s size, although it had low efficacy. ",
      "research_paper": "assets/pdfs/65d96623dbb39e8adc85aea5d1c4803112a66d13.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/Ian-Inspirit-AI-Code/DataProcessing",
      "graphic_link": "https://cdn.britannica.com/83/232283-050-7194EBA7/Tsunami-hits-sea-Miyako-Japan-March-11-2011.jpg",
      "headline": "Many tsunamis have enough energy to cause a large amount of human death, as well as trillions of dollars in damage repair. The problem arises on how to better prepare for these natural disasters and limit its damage, especially in countries such as Japan that are frequented by tsunamis.",
      "project_id": "ccfcd2d500fd177b3719b2a5ffec20d8a7b703ed",
      "expand": true,
      "tags": [
        "Logistic Regression",
        "Earth Systems",
        "Natural Disasters"
      ],
      "related_proj": [
        95,
        121,
        54
      ]
    },
    {
      "student_name": "Animish J.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Optimizing Skin Cancer Classifiers By Applying Multiplicative Weight Update Into A Mobile Application",
      "project_desc": "This study deals with using Artificial Intelligence and Machine Learning in order to create models that can diagnose skin cancer using images. Many models have been created in the past using different algorithms and methods. In this study, the multiplicative weight update method is used in order to take the predictions of multiple models in order to try and acquire more accurate predictions. In this study, a Logistic Regression, CNN, and SVC model are used. These models are sent images of skin cancer from the ISIC-Archive and they try to recognize patterns in order to categorize new images. These models are all then sent to a multiplicative weight update algorithm which takes into account the precision and accuracy of each model through each successive guess in order to add weight to their guess. These guesses are then added together in order to try and find the correct predictions. Using Multiplicative Weight Update, the model received an accuracy of 74.69%. The conclusion was made that using an SVC  model would be the best option for this problem rather than a Multiplicative Weight Update system. \n",
      "research_paper": "assets/pdfs/0e313744bc3f9c40120b0cc546ee54b273648f23.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://res2.yourwebsite.life/res/5f7f15d2dd43c80022ec7b78/5f7fa95496628c00210b889e_optimized.webp",
      "headline": "Skin cancer is slowly becoming the most common and frequent type of cancer in the world. In the United States alone, research suggests that skin cancer is the most common cancer in and that one in every five people will acquire skin cancer at some point throughout their lifetime.",
      "project_id": "a11279d957e8a60c671b45c11106105670960e49",
      "expand": true,
      "tags": [
        "Skin Cancer",
        "Diseases and Health Conditions",
        "App Development",
        "Cancer"
      ],
      "related_proj": [
        152,
        174,
        135
      ]
    },
    {
      "student_name": "Matthew N.",
      "mentor_name": "Sarthak Kanodia",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment"
      ],
      "project_title": "A Comprehensive Study of Machine Learning Models for the Prediction of pH in Fairfax County",
      "project_desc": "Water quality is drastically declining as bodies of water \nbecome more and more acidic due to global warming. \nBecause of this, it is crucial to understand what factors \ndetermine pH, and how to predict pH values. In this study, \nwe determine which features are most important in \ndetermining pH, and also what combination of features and \nmachine learning models give us the best accuracy in the \nprediction of pH values. We demonstrate that pH values \nfrom the previous year and the percentage of rehabilitated \nsanitary sewer lines are the most important features in \ndetermining pH. We also demonstrate that Decision Tree \nModels and Random Forest Models perform the best across \nall features and feature subsets.",
      "research_paper": "assets/pdfs/c423ee893b24bf4a7d44b90f31f8fa91dfd74289.pdf",
      "project_yr": "2023",
      "project_quarter": "Spring",
      "graphic_link": "https://www.epa.gov/system/files/styles/gallery_medium/private/images/2023-06/Get%20Data%20Hand%20in%20water%20with%20drops.jpg?h=64cd3a75&itok=aaMldG5n",
      "headline": "Water quality is drastically declining as bodies of water become more and more acidic due to global warming. Because of this, it is crucial to understand what factors determine pH, and how to predict pH values. In this study, we determine which features are most important in determining pH, and also what combination of features and machine learning models give us the best accuracy in the prediction of pH values. We demonstrate that pH values from the previous year and the percentage of rehabilitated sanitary sewer lines are the most important features in determining pH. We also demonstrate that Decision Tree Models and Random Forest Models perform the best across all features and feature subsets.",
      "project_id": "e99ba3f040d469c86d17bed3f2bdd03a9f653c89",
      "expand": true,
      "tags": [
        "Earth Systems",
        "Chemistry",
        "Water"
      ],
      "related_proj": [
        41,
        37,
        107
      ]
    },
    {
      "student_name": "Jaida G.",
      "mentor_name": "Odysseas Drosis",
      "mentor_title": "PhD Candidate in Computer Science, Masters in Computer Science Alum from Cornell",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Skin Cancer Detection ",
      "project_desc": "The goal of this research project is to predict whether or not a patient has skin cancer through a machine learning model that is developed from an image dataset. Skin cancer is extremely dangerous, as over 9500 people in the US are diagnosed with it daily. If detected early, patients will have a more likely chance of survival. I tested an MLP Classifier, Decision Tree Regressor, a Logistic Regression Model, and a KNN Model to compare various results and ultimately determine the best accuracy. The MLP Classifier had a 74.5% accuracy, the Decision Tree Regressor had a 74.1% accuracy, the Logistic Regression Model had a 68.8% accuracy, and the KNN Model had a 74.6% accuracy (all testing). We can see that the MLP Classifier, Decision Tree Regressor, and the KNN Model had around the same accuracy while outperforming the Logistic Regression Model. However, when comparing training data, there seems to be a large overfitting problem with most of the models. ",
      "research_paper": "assets/pdfs/5287c08277891732a6dab1b489416e15849b62e9.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.cnet.com/a/img/resize/003ff7bf85a1a57a2a57fcf043b9160f6a47a9d7/hub/2019/06/26/b2b042fc-4ae0-4a80-9ae7-11cd7eba0a59/skinvision.jpg?auto=webp&width=1200",
      "headline": "The goal of this research project is to predict whether or not a patient has skin cancer through a machine learning model that is developed from an image dataset. Skin cancer is extremely dangerous, as over 9500 people in the US are diagnosed with it daily. If detected early, patients will have a more likely chance of survival. I tested an MLP Classifier, Decision Tree Regressor, a Logistic Regression Model, and a KNN Model to compare various results and ultimately determine the best accuracy. The MLP Classifier had a 74.5% accuracy, the Decision Tree Regressor had a 74.1% accuracy, the Logistic Regression Model had a 68.8% accuracy, and the KNN Model had a 74.6% accuracy (all testing). We can see that the MLP Classifier, Decision Tree Regressor, and the KNN Model had around the same accuracy while outperforming the Logistic Regression Model. However, when comparing training data, there seems to be a large overfitting problem with most of the models.",
      "project_id": "14fe693153a6a674ce2cf5f7b2028b54c56e3b8a",
      "expand": false,
      "tags": [
        "Cancer",
        "Diseases and Health Conditions",
        "Skin Cancer"
      ],
      "related_proj": [
        8,
        64,
        108
      ]
    },
    {
      "student_name": "Giuliano T.",
      "mentor_name": "Sophia Barton",
      "mentor_title": "Computer Science MS from Stanford",
      "mentor_image": "assets/images/mentor_imgs/56c2b2fb8dc613f5a0a7ae222a5ee595734e2950.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "The Motions of Forest Fires",
      "project_desc": "Can I make an AI model that can predict how forest fires will move? I am trying to take data about the environment and a current fire and predict where the fire will go / if it will grow or shrink. Forest fires are burning more and more each year, as evidenced by the record breaking fires in Australia in 2020. The top ten years with the largest annual acreage burnt have all happened since 2004, and forest fires are now causing 7.4 million more acres of tree cover loss per year since 2001.\nThis project would hopefully let firefighters predict where the fire is going to prevent it. The data I used was a graph of where the fire was and other features such as the surrounding wind speed, terrain, plant content, etc, and then the label comes from another graph showing where the fire moved to. I tried several different models, namely a Multi Layer Perceptron (or Neural Net), Logistic Regression and Linear Regression before settling on a CNN that predicts if each square kilometer will have a fire or not. I then was unable to get this to work so I settled on making a model that predicts how dangerous the fire will be. I was eventually able to get it to predict this with 60% accuracy. The major conclusion I had with this problem is that it is very tough but it seems like there is a way to predict where a forest fire will go. I would like to keep researching to see if I can find out how.\n",
      "research_paper": "assets/pdfs/04bd09edf392cd60f2114e510cfa166a5b7a4990.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://insideclimatenews.org/wp-content/uploads/2023/03/wildfire_thibaud-mortiz-afp-getty-scaled.jpg",
      "headline": "Can I make an AI model that can predict how forest fires will move? I am trying to take data about the environment and a current fire and predict where the fire will go / if it will grow or shrink. ",
      "project_id": "e129d1129e9f9dc7f421566b7409f57da0680cff",
      "expand": false,
      "tags": [
        "Earth Systems",
        "Forest Fires",
        "Logistic Regression"
      ],
      "related_proj": [
        41,
        68,
        35
      ]
    },
    {
      "student_name": "Aryaman R.",
      "mentor_name": "Yuan Lee",
      "mentor_title": "EECS MEng from MIT",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Approaches to fraud detection on credit card transactions using artificial intelligence methods",
      "project_desc": "In this paper, we study the problem of detecting fraudulent credit card transactions. We select the most relevant features using a heuristic approach, and fit three different model classes to a simulated dataset: Logistic Regression, Random Forests and Gradient Boosting Machines. We find that hyperparameter tuning has a big impact on the precision and recall of our classifiers. We also find that of the three classes, Gradient Boosting Machines were the best-performing model class, achieving 83% precision and 64% recall on unseen data.",
      "research_paper": "assets/pdfs/0cdd8a9fe0af140bea1a08e1b3255da866789f3b.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.incharge.org/wp-content/uploads/2019/12/credit-card-fraud.png",
      "headline": "In this paper, we study the problem of detecting fraudulent credit card transactions. We select the most relevant features using a heuristic approach, and fit three different model classes to a simulated dataset: Logistic Regression, Random Forests and Gradient Boosting Machines. We find that hyperparameter tuning has a big impact on the precision and recall of our classifiers. We also find that of the three classes, Gradient Boosting Machines were the best-performing model class, achieving 83% precision and 64% recall on unseen data.",
      "project_id": "a56ab8773f5c9ee799a933a83bb78a4f62887aa4",
      "expand": true,
      "tags": [
        "Economics",
        "Consumer Behavior",
        "Fraud"
      ],
      "related_proj": [
        116,
        186,
        88
      ]
    },
    {
      "student_name": "Alexander F.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "PoliSci and Law"
      ],
      "project_title": "Predicting Recidivism in the United States Criminal Justice System",
      "project_desc": "We investigated how artificial intelligence creates bias in the US criminal justice system. Recidivism prediction algorithms used in courtrooms have been shown to produce unnecessarily harsh risk assessment scores against Black individuals, leading them to have more strict bail guidelines. This contributes to a greater narrative against Black Americans. In order to explore how different algorithms might be biased in predicting recidivism, we used 3 different classification models to predict whether individuals in the Virginia Felony Database dataset would recidivate. We tested three different (optimized) machine learning algorithms — Logistic Regression, Random Forest, and k-nearest-neighbors and compared the resulting confusion matrices to determine which was the most effective. The final Logistic Regression model had a final accuracy of 70.44%; the Random Forest algorithm had the best at 75.64% and K-nearest-neighbors had 72.24%. All algorithms had a greater false positive (false arrest prediction) for white individuals, likely due to the dataset containing many examples of white criminals. In our model white defendants were discriminated against because of data bias, which demonstrates how algorithms can treat different racial groups differently depending on the data that goes in. Given historical bias, we can see how this same pattern could occur in the opposite direction favoring white individuals more frequently in other datasets.",
      "research_paper": "assets/pdfs/6d7ebe606bfd552e39df982e18bff9413d4021f0.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.povertyactionlab.org/media/image/20170505-recidivismjpg",
      "headline": "There are real-life effects of machine bias in courtrooms. American lives are completely changed based on criminal sentencing, which can mean the difference between rehabilitation and recidivism. This issue is urgent and finding a solution to ensure equality is imperative.",
      "project_id": "3b1e19d91a523f901204bc8995c84da97f5085cd",
      "expand": false,
      "tags": [
        "Criminal Justice",
        "Recidivism",
        "Law and Order"
      ],
      "related_proj": [
        100,
        72,
        79
      ]
    },
    {
      "student_name": "Milo B.",
      "mentor_name": "Samuel Kwong",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Classification of Exon and Intron Boundaries",
      "project_desc": "The goal was to accurately classify exon and intron boundaries based on DNA sequences. Scientists can learn more about proteins if divisions between exons and introns are clear. We used multiple different machine learning approaches including a logistic regression model, a multilayer perceptron, a LSTM, and a model that included a LSTM, autoencoder and a multilayer perceptron. LSTMs performed well, pointing to the idea that order of nucleotides is important when classifying DNA sequences.",
      "research_paper": "assets/pdfs/5e52f405b21ba1d699d6791a0af9798ec16b4a7d.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.news-medical.net/image.axd?picture=2018%2F11%2Fshutterstock_231128206.jpg",
      "headline": "The goal was to accurately classify exon and intron boundaries based on DNA sequences. Scientists can learn more about proteins if divisions between exons and introns are clear. We used multiple different machine learning approaches including a logistic regression model, a multilayer perceptron, a LSTM, and a model that included a LSTM, autoencoder and a multilayer perceptron. LSTMs performed well, pointing to the idea that order of nucleotides is important when classifying DNA sequences.",
      "project_id": "135ecb38979e957454d852cfe9e34cd39eee2ec6",
      "expand": false,
      "tags": [
        "Biology",
        "DNA",
        "LSTMs",
        "Logistic Regression"
      ],
      "related_proj": [
        31,
        171,
        103
      ]
    },
    {
      "student_name": "Marco Y.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Sports and Motion Sciences"
      ],
      "project_title": "Application of AI to Tennis Match Footage Transcription",
      "project_desc": "One of the best ways for tennis players to improve their game is to\nrecord and watch their own match footage, find patterns in the points\nthey win and lose, and practice based on these realizations. However,\nwatching match footage and documenting each point shot by shot is a\nvery time-consuming process. This paper investigates an AI approach to\ntranscribing tennis match footage, combining a deep convolution neural\nnetwork (YOLOv4), a pose estimation model (Movenet), and a long short-\nterm memory (LSTM) deep neural network. Looking at a transcript of\neach point will be far more efficient than watching entire match footage\nfor a player to understand how they are losing and winning and analyze\npatterns in their game. The LSTM model in this project achieved accura-\ncies of 73.33% and 79.31% when classifying shot type (forehand, forehand\nvolley, forehand slice, backhand, backhand volley, backhand slice, over-\nhead/smash, and serve) for players on the close side and opposite side\nof the net, respectively, and 55.17% and 60.00% when classifying the di-\nrection of a shot (cross-court, down the line, down the middle, inside in,\ninside out, out wide, down the t, and body) for players on the close side\nand opposite side of the net, respectively.\n1 Introduction\n1.1 Motivation\nAs a competitive junior tennis player, I sit down with a notebook after the match\nand watch my game footage. Documenting what happens in each point shot by\nshot is critical for players trying to improve their game. By analyzing match\nfootage a player can realize specifically what they need to work on to play better\nduring their next tournament, whether that is a shot, pattern, or tendency. For\nexample, if a player notices that he or she is losing a lot of points with long,\nforehand cross-court rallies, he or she can train the forehand cross-court shots.\nTennis match transcription is not just important to players at the amateur and\nprofessional level who are trying to improve, but also to research in the field\nof tennis analytics, as generating information for tennis matches is the first\n1\n",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://img.olympicchannel.com/images/image/private/t_s_pog_staticContent_hero_xl_2x/f_auto/primary/weyzoz6o394sxfkbest9",
      "headline": "One of the best ways for tennis players to improve their game is to record and watch their own match footage, find patterns in the points they win and lose, and practice based on these realizations. However, watching match footage and documenting each point shot by shot is a very time-consuming process. This paper investigates an AI approach to transcribing tennis match footage, combining a deep convolution neural network (YOLOv4), a pose estimation model (Movenet), and a long short- term memory (LSTM) deep neural network. Looking at a transcript of each point will be far more efficient than watching entire match footage for a player to understand how they are losing and winning and analyze patterns in their game. The LSTM model in this project achieved accura- cies of 73.33% and 79.31% when classifying shot type (forehand, forehand volley, forehand slice, backhand, backhand volley, backhand slice, over- head/smash, and serve) for players on the close side and opposite side of the net, respectively, and 55.17% and 60.00% when classifying the di- rection of a shot (cross-court, down the line, down the middle, inside in, inside out, out wide, down the t, and body) for players on the close side and opposite side of the net, respectively.",
      "project_id": "97a5b129fe54fbe0de6a6e5cf0be48752691b5df",
      "expand": false,
      "tags": [
        "Sports",
        "LSTMs"
      ],
      "related_proj": [
        4,
        112,
        11
      ]
    },
    {
      "student_name": "Deetya N.",
      "mentor_name": "Linda Banh",
      "mentor_title": "Electrical Engineering MS from Stanford with professional AI Certificate, Engineer at Microsoft",
      "mentor_image": "assets/images/mentor_imgs/3c9d0f6a41bcd9e11d20529cfd74449f29a85bb9.png",
      "student_image": "assets/images/student_imgs/7da68d2a7ef90411627e3b7022a256f23042e193.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Predicting Drug-Drug Interaction Severity Using Network Characteristics",
      "project_desc": "Drug-drug interactions (DDIs), which can add to or diminish the effect of one drug or impact the metabolism of one drug, have harmful effects on health in patients that take multiple drugs. Testing for DDIs is slow and costly, so computational models have recently been used to predict them. Network information is useful in describing a drug’s known interactions and mechanisms to determine whether two drugs could be interacting. This research explores various machine learning models to predict the severity of unknown interactions of existing drugs (major, minor, or moderate), using the DDInter database. The best-performing model, a multi-layer perceptron model which concatenates two drugs’ embeddings, had an accuracy of about 92%, average precision of about 87.1% (89.4% for the major category, 78.5% for minor, and 93.5% for moderate) and average recall of 83.3% (86.4% for major, 70.1% for minor, and 95.3% for moderate). Future work in this area includes refining models to increase precision and recall, obtaining a more balanced dataset, predicting the mode of interaction, and using other drug characteristics. The code for this project can be found here, along with a Streamlit deployment of the model: https://github.com/deetyabn/DDI_severity_prediction. ",
      "research_paper": "assets/pdfs/4b20dff436fd43b9c03333720613b533d6843582.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/deetyabn/DDI_severity_prediction",
      "project_webpage": "https://colab.research.google.com/drive/1ODK9UsFpHrRFA54kDgmITYIGa9YSD0sd?usp=sharing",
      "graphic_link": "https://www.grxstatic.com/4f3rgqwzdznj/4YlSpmw5yn50QkFRPQV5wL/af16db25565cbbbb551e7cbb43a0adb1/spilling_pills_various_pink_background-1174440156.jpg?format=pjpg&auto=webp&width=704",
      "headline": "Drug-drug interactions (DDIs), which can add to or diminish the effect of one drug or impact the metabolism of one drug, have harmful effects on health in patients that take multiple drugs. Testing for DDIs is slow and costly, so computational models have recently been used to predict them. Network information is useful in describing a drug’s known interactions and mechanisms to determine whether two drugs could be interacting. This research explores various machine learning models to predict the severity of unknown interactions of existing drugs (major, minor, or moderate), using the DDInter database.",
      "project_id": "bfe1af68283bedebfde2328478d0a265c3429119",
      "expand": false,
      "tags": [
        "Drugs",
        "Biology",
        "Web Deployment"
      ],
      "related_proj": [
        73,
        97,
        153
      ]
    },
    {
      "student_name": "Leo R.",
      "mentor_name": "Roger Jin",
      "mentor_title": "Computer Science MS at MIT, ML researcher at Apple",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": "SaShiMi: Adapted for Google Colab",
      "project_desc": "Adapting code and how it runs to allow SaShiMi to run on Google's resources",
      "research_paper": "assets/pdfs/069b465e1d7b252a5b98f92d617dad97b3cae7b0.pdf",
      "project_yr": "2022",
      "project_quarter": "Spring",
      "graphic_link": "https://encrypted-tbn3.gstatic.com/images?q=tbn:ANd9GcSYY6n1fYadsaidY0V8tPq6vomWkb8qu9y5lME7SKy7Pe4UKDn9",
      "headline": "In this project, I convert SaShiMi, a music generation software, into something more resource-friendly.",
      "project_id": "5e1b2bfaa9c8506d5425ef0bdff1944ddaff00c7",
      "expand": false,
      "tags": [
        "Music",
        "Optimization",
        "Generative AI"
      ],
      "related_proj": [
        146,
        84,
        87
      ]
    },
    {
      "student_name": "Khondoker F.",
      "mentor_name": "Barbie Duckworth",
      "mentor_title": "Computer Science BS Alumni MIT, prior Software Engineer at Facebook and eBay",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/73eda021247af8b9c6e12156f6daa7cb38c3f986.png",
      "domains": [
        "Engineering",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "DeepSolar Bangladesh: A Novel Convolutional Neural Network (CNN) Architecture for the Detection of Solar Panels from Low Resolution Satellite Imagery in Developing Countries",
      "project_desc": "Due to its environmental benefits and decreasing costs, the supply of solar energy is growing at an accelerating pace globally. However, the decentralised nature of solar makes it difficult to keep track of the different photovoltaic (PV) systems deployed across a country. There is a critical need for highly accurate, comprehensive national databases of solar systems, which would allow policymakers, researchers, and the government to study socioeconomic trends in solar deployment. Manual surveys have shown to be inaccurate. The 2018 DeepSolar study by Yang et. al developed a deep-learning framework and national solar deployment database for the US using high-quality satellite imagery, which proved to be a much more efficient and accurate approach. However, satellite imagery in developing countries such as Bangladesh is of much lower resolution and quality, and performed poorly with the original DeepSolar model by Yang et. al. Our study highlights the implementation of a novel convolutional neural network (CNN) in detecting solar panels through low resolution Google Static Maps API satellite imagery data. The model was trained over 500 epochs and had 49,859,906 parameters, and classified image samples as positive (indicating presence of solar panels) or negative (absence of solar panels). Our accuracy was 86.49%, F1 score was 86.49%, precision was 91.95%, and recall score was 81.63%, which are comparable scores to the original DeepSolar CNN that was trained on much higher quality data. This was the first CNN to detect solar panels using low-resolution satellite imagery (which is usually the only option for developing countries), and showed to be highly accurate and computationally efficient. Future plans include gathering funding to be able to purchase Google API satellite imagery to be able to cover all of Bangladesh to create a comprehensive public national database, as well as expanding to other developing countries.",
      "research_paper": "assets/pdfs/868b76e02b4b9ebab1a8e8335b806122bcb679ce.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.pv-magazine.com/wp-content/uploads/2021/04/Bild1-1-1200x563.jpg",
      "headline": "Due to its environmental benefits and decreasing costs, the supply of solar energy is growing at an accelerating pace globally. However, the decentralised nature of solar makes it difficult to keep track of the different photovoltaic (PV) systems deployed across a country. There is a critical need for highly accurate, comprehensive national databases of solar systems, which would allow policymakers, researchers, and the government to study socioeconomic trends in solar deployment. Manual surveys have shown to be inaccurate. The 2018 DeepSolar study by Yang et. al developed a deep-learning framework and national solar deployment database for the US using high-quality satellite imagery, which proved to be a much more efficient and accurate approach. However, satellite imagery in developing countries such as Bangladesh is of much lower resolution and quality, and performed poorly with the original DeepSolar model by Yang et. al. Our study highlights the implementation of a novel convolutional neural network (CNN) in detecting solar panels through low resolution Google Static Maps API satellite imagery data.",
      "project_id": "4777f1633b5a64e6b806fbb637962cdb73afde74",
      "expand": true,
      "tags": [
        "Earth Systems",
        "Renewable Energy",
        "Satellite Imagery",
        "Solar Power"
      ],
      "related_proj": [
        30,
        159,
        141
      ]
    },
    {
      "student_name": "James W.",
      "mentor_name": "Aidan Donaghey",
      "mentor_title": "Physics Yale Alumni, Researcher in Bioinformatics and Astrophysics",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/640021b1c2e48914c787d13228b2d9c18aada8c7.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Molecular Analysis of Stellar Clusters Final Paper",
      "project_desc": "WIth the diversity of research methods and measurement instruments found in astronomy, it is often difficult to draw direct comparisons between conclusions due to differences in both kind and quality of measurements.   By using a machine learning algorithm we have been able to predict measurements that may not have been gathered, notably , the iron content of a star based upon temperature or gravity readings. The problem has not been answered before with this data set, and the scientific community does not have a large amount of research that has pertained to this specific APOGEE data release. By addressing this previously unaddressed problem, we can advance the scientific community with a new contribution that allows scientists of the future to gain additional data in conditions where such data may not be so readily available. We have approached the question by using various machine learning algorithms to discover correlations between the data points we were given, creating a predictive model to produce a numerical result. Gravity and temperature do not present a direct linear relationship to the iron content of a star. However, as their data points tend to cluster together, we can utilize the KNeighborsRegressor to find correlations in the way that these points have, and produce a numerical prediction. Throughout testing we’ve found that the KNeighborsRegressor does better on all accounts in comparison to a large variety of other regression models such as the GaussianProcessRegressor. Considering the results of the Mean Absolute Percentage Error function upon the results of both of these, we can note how the LinearRegressor, mathematically, does marginally better than the KNeighborsRegressor, at a value of 1.13% of average error in comparison to 2.53%. ",
      "research_paper": "assets/pdfs/52f277b11a08398d811fce53cf85e6dbfc8a7f4e.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.astronomytrek.com/wp-content/uploads/2017/03/Jewel-Box-star-cluster-1280x720.jpg",
      "headline": "By testing a multitude of models from the python package scikit-learn upon the APOGEE dataset, we’ve managed to produce a resultant product that can predict a star’s iron concentration depending upon its gravitational pull (LOGG) value or effective temperature (TEFF) value.",
      "project_id": "28be4b1740f489c7b5116dfe0095211e74b1cfea",
      "expand": false,
      "tags": [
        "Star Clusters",
        "Astronomy"
      ],
      "related_proj": [
        99,
        166,
        3
      ]
    },
    {
      "student_name": "Alexander K.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/student_imgs/88f1d057d0384144cf8e00a10d34c57747e2bdf5.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Machine Learning Algorithms for Consumer Plastics Identification and Sorting",
      "project_desc": "Seeking a solution to maintain plastic recycling costs while increasing the output of reusable material, this paper poses the question “what type of machine learning algorithm is most suitable for a plastic identification system in consumers’ homes?” It investigates five total machine learning algorithms to determine which one best balances accuracy, management of resources, and time efficiency, ultimately arriving at the conclusion that a Support Vector Machine that uses a Polynomial Kernel is the best algorithm and serves to demonstrate that algorithms such as the ones analyzed have become advanced enough for more efficient, AI driven systems to replace those of the status quo.",
      "research_paper": "assets/pdfs/77ff4191da8256dcdeac81ff046d787fa8c61cea.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://waste-management-world.com/imager/media/wasteManagementWorld/2231185/AdobeStock_320996375_428fd902f4247199467725e7eccf1673.jpeg",
      "headline": "Seeking a solution to maintain plastic recycling costs while increasing the output of reusable material, this paper poses the question “what type of machine learning algorithm is most suitable for a plastic identification system in consumers’ homes?” It investigates five total machine learning algorithms to determine which one best balances accuracy, management of resources, and time efficiency, ultimately arriving at the conclusion that a Support Vector Machine that uses a Polynomial Kernel is the best algorithm and serves to demonstrate that algorithms such as the ones analyzed have become advanced enough for more efficient, AI driven systems to replace those of the status quo.",
      "project_id": "419f2aa1abd6d04e1bb0ee5d9ed0f798c461d94a",
      "expand": true,
      "tags": [
        "Earth Systems",
        "Recycling"
      ],
      "related_proj": [
        27,
        23,
        172
      ]
    },
    {
      "student_name": "Nuo W.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Attention LSTMs in Multimodal Models: A holistic approach to predicting COVID infection trends",
      "project_desc": "COVID-19 and its infection trends are a hot topic in the AI space due to the seeming applicability of AI, which could help with case preventions and medical resource allocations. However, too often people focus on a predetermined area and scale, overlooking the fact that COVID doesn’t flow into or out of just one area but instead spreads everywhere simultaneously. Thus, we propose novel approaches to process the infection rates of all states in the United States as a whole while taking into account the interconnectivity between states through domestic flights. We introduce 1) the graph attention LSTM for analyzing domestic flights across states, 2) the image attention LSTM for understanding infection rates from United States COVID heat maps, and 3) the attention bottleneck mid fusion model that integrates graph and image attention LSTMs. Through tests that require simultaneous predictions of the infection rates in all US states, we compare the performances of our methods with those of classic sequence-processing methods and achieve significantly more accurate results.",
      "research_paper": "assets/pdfs/60c004c21f0b82bf2ddb65dbb9325928468fd194.pdf",
      "project_yr": "2022",
      "project_quarter": "Spring",
      "github": "https://github.com/NuoWenLei/Attention-LSTMs-in-Multimodal-Models",
      "project_webpage": "https://nuowenlei-attention-lstms-in-multimoda-model-comparison-hoq2rx.streamlitapp.com/",
      "graphic_link": "https://www.news-medical.net/images/news/ImageForNews_722926_16611632963032177.jpg",
      "headline": "In our search for a way to simultaneously predict all state-level COVID infection rates in the United States with COVID heat maps and domestic flight graphs, we propose novel methods of processing graph and image sequences with attention-based LSTM layers as well as evaluate the effectiveness of different multimodal fusion techniques.",
      "project_id": "c030fe0fd0857c2c8640e4879ee074ef767e49f6",
      "expand": true,
      "tags": [
        "LSTMs",
        "COVID-19",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        126,
        50,
        174
      ]
    },
    {
      "student_name": "Ram N.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "One Class Classification for Overdose Death Detection",
      "project_desc": "This paper describes a model intended to predict drug overdose deaths given a certain amount of information. This was done by analyzing and formatting data of drug overdose deaths in the US during 2012 to 2018. The data was then put into a One Class Classification model which would be able to predict the likelihood of a drug overdose death. Three different models were used returning an f1 score of 0.836, 0.849, and 0.85 respectively.  ",
      "research_paper": "assets/pdfs/82aa2b4124289b768db55bc41ff09bfb98e08656.pdf",
      "project_yr": "2022",
      "project_quarter": "Spring",
      "graphic_link": "https://people.com/thmb/mp3LBpiYOjPq1MgwHKJBgd_ri70=/1500x1000/filters:fill(auto,1)/drug-overdose-2-6599dce6b736463b825c1de98aefaec2.jpg",
      "headline": "In the past year alone there were an estimated 107,622 drug overdose deaths in just the United States. With such an incredible amount of deaths from just this, I thought it would be beneficial to create a model to predict who is at risk of drug overdose deaths.",
      "project_id": "d2fa188f0c1fc416dab710c7365bd445f6bd6018",
      "expand": false,
      "tags": [
        "Drug Overdose",
        "Drugs"
      ],
      "related_proj": [
        184,
        155,
        104
      ]
    },
    {
      "student_name": "Arianna H.",
      "mentor_name": "Matteo Santamaria",
      "mentor_title": "Mathematical & Computational Engineering MS from Stanford, Data Scientist at infima",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Journalism"
      ],
      "project_title": "Biased News Detection Using Artificial Intelligence",
      "project_desc": "Biased news has become especially problematic in modern times. In this paper, we investigate the abilities of various machine learning models to correctly identify biased news. Our results indicate that artificial intelligence is very capable of doing this detection. In particular, both logistic regression and K-nearest neighbor classifier performed rather well in cross validation. ",
      "research_paper": "assets/pdfs/cdd9c0c92540083d70801edb3b5de1666857fc95.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.usnews.com/object/image/00000185-f52e-d39b-a98f-ff6e7bdf0000/gettyimages-1239793196.jpg?update-time=1674855937422&size=responsive640",
      "headline": "Biased news has become especially problematic in modern times. In this paper, we investigate the abilities of various machine learning models to correctly identify biased news. Our results indicate that artificial intelligence is very capable of doing this detection. In particular, both logistic regression and K-nearest neighbor classifier performed rather well in cross validation. ",
      "project_id": "75980cadee1c8968603f2a45662662cc2fe31db8",
      "expand": false,
      "tags": [
        "News",
        "Fake News",
        "Natural Language Processing"
      ],
      "related_proj": [
        87,
        81,
        154
      ]
    },
    {
      "student_name": "Bwohan W.",
      "mentor_name": "Barbie Duckworth",
      "mentor_title": "Computer Science BS Alumni MIT, prior Software Engineer at Facebook and eBay",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/7198ef8fe690ff8d581bb59073aee05e03cf28de.png",
      "domains": [
        "Healthcare and Biology",
        "Psychology"
      ],
      "project_title": "Training a Machine Learning Model to Recognize Signs of Depression",
      "project_desc": "Depression is a common mental disorder yet can severely damage your everyday habits and health. It is also difficult to diagnose without a professional assessment, so it can be hard to identify whether or not you have it. However, some everyday habits, and changes in those habits, can be indicative of future depressive episodes. Pinpointing those features, I trained a KNN and decision tree based models to predict whether or not someone would have a depressive episode in the future. The best performing model was a KNN model which had an accuracy of 70%, with an F1 score of 69%. These statistics, while not insignificant, still need to be improved on before they can be used as a reliable predictor of depression.",
      "project_yr": "2022",
      "project_quarter": "Spring",
      "graphic_link": "https://images.everydayhealth.com/images/how-to-cope-with-anxiety-and-depression-722x406.jpg",
      "headline": "Many people have not had a professional diagnosis, and could have depression without being aware of it. To attempt to answer this question, I used a person’s daily habits and how they change, and attempt to predict if they will experience a depressive episode in the future.",
      "project_id": "668fe721f198b9326694fc2f33856dcef5f31f81",
      "expand": false,
      "tags": [
        "Mental Health",
        "Depression",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        32,
        10,
        137
      ]
    },
    {
      "student_name": "Rohan S.",
      "mentor_name": "Sriram Hathwar",
      "mentor_title": "Princeton Alum, Bioinformatics Researcher",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Brain Tumor Classification",
      "project_desc": "Early classification and diagnosis of Brain Tumors is essential for providing the right treatment to a patient. It is crucial to get treatment as soon as possible because the survival rate for someone with an untreated brain tumor can range from as low as 3 months to as high as 5 years, which is why it is extremely important to get treatment as soon as possible. In this project, we used brain tumor MRI images to classify them into these 4 different categories: glioma, meningioma, pituitary, and no tumor. With the use of baseline and deep learning models, the deep learning models demonstrated a significantly higher performance due to their ability to analyze images. The model with the highest accuracy was the MobileNet model, a pre-trained transfer learning model trained on 5,608 images. This model yielded  a validation accuracy of 98.24%. Using metrics including Kappa cohen score, precision, and recall, we validated the performance of the machine learning model.   We deployed the MobileNet model to a web app using Streamlit, where users submit MRI images and receive diagnoses of tumor class.  We found that the model performed very well while utilizing the web app, indicating that it is safe to be used. However, since we only have 4 classes and there are over 150 total types of brain tumors, it could easily get a diagnosis wrong if it is not in one of these 4 classes.",
      "research_paper": "assets/pdfs/db2e298b8e470492951e248a761a9b01d373c6b4.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/rohansangal1/Brain-Tumor-Classification",
      "graphic_link": "https://prod-images-static.radiopaedia.org/images/19319983/e8b196f5268ae6dc7f6915cbb48c88_big_gallery.jpeg",
      "headline": "Early classification and diagnosis of Brain Tumors are essential for providing the right treatment to a patient. It is crucial to get treatment as soon as possible because the survival rate for someone with an untreated brain tumor can range from as low as 3 months to as high as 5 years. In this project, we classified brain tumor images into 4 categories: glioma, meningioma, pituitary, and no tumor. With the use of baseline and deep learning models, the deep learning models demonstrated a significantly higher performance due to their ability to analyze images. The model with the highest accuracy was the MobileNet, a pre-trained transfer learning model trained on 5,608 images. This model yielded a validation accuracy of 98.24%. Using metrics including Kappa cohen score, precision, and recall, we validated the machine learning model's performance. We deployed the MobileNet model to a web app using Streamlit, where users submit MRI images and receive diagnoses of tumor class. We found that the model performed very well while utilizing the web app, indicating that it is safe to be used. However, since we only have 4 classes and there are over 150 total types of brain tumors, it could easily get a diagnosis wrong if it is not in one of these 4 classes.",
      "project_id": "3bdffe331f52081e0f2024f17e079d5ec940cc41",
      "expand": false,
      "tags": [
        "Cancer",
        "Brain Cancer",
        "Neuroscience"
      ],
      "related_proj": [
        108,
        117,
        156
      ]
    },
    {
      "student_name": "Isaac A.",
      "mentor_name": "Aidan Donaghey",
      "mentor_title": "Physics Yale Alumni, Researcher in Bioinformatics and Astrophysics",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics"
      ],
      "project_title": "An Investigation Into Applications of Machine Learning Algorithms on Solar Flare Data and Distance Prediction",
      "project_desc": "A solar flare is an intense localised eruption of electromagnetic radiation in the sun’s atmosphere. A solar flare is typically accompanied with a coronal mass ejection (CME), in which a highly magnetic plasma is released from the sun’s corona into the heliosphere. CMEs are capable of reaching earth and colliding with the earth’s magnetic field, causing dangerous geomagnetic storms that can start fires and cause power outages. It is imperative that flares are studied so that actions can be taken to mitigate the effect of geomagnetic storms. Our data was sourced from the RHESSI telescope which consisted of almost 100,000 entries. It was shown in Fletcher’s paper that there exists a correlation between radial distance and the energy of a flare. We began with classifying solar flares based on their radial distances and then applying machine learning models like logistic regression, KNN classification, Decision Trees and MLP Classification and closely examining the ones that gave the highest accuracy. Our different machine learning models showed that there existed a pattern in the phenomena of solar flares and our AI could be used to predict them accurately. Most notably, the decision tree model had an accuracy of 99.84%.\n",
      "research_paper": "assets/pdfs/df76dbf0c48726c884f7cde63359a8e2ebcd47ae.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.slashgear.com/img/gallery/heres-how-solar-storms-can-impact-the-earth/solar-flares-and-earth-1663205183.jpg",
      "headline": "The current method that NASA uses for flare prediction involves studying various solar cycles that range from 11 days to 80 years. However, there are far too many factors to consider using this method of prediction and the forecasts are often wrong.",
      "project_id": "3da55ed927a10c8aac6d7150c913ad187e288721",
      "expand": true,
      "tags": [
        "Astronomy",
        "Solar Flares"
      ],
      "related_proj": [
        20,
        36,
        129
      ]
    },
    {
      "student_name": "Samantha K.",
      "mentor_name": "Akshay Jagadeesh",
      "mentor_title": "Computational Neuroscience PhD from Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Psychology",
        "Healthcare and Biology"
      ],
      "project_title": "The Impacts of Child-Mentor Relationships on Child Mental Health",
      "project_desc": "With mental health being a field that is becoming progressively more familiar and researched in depth, an important subject of interest that should be addressed is child mental health, and the different factors that can influence it. In this paper, we use a dataset from the Substance Abuse and Mental Health Data Archive to analyze the impacts of child-mentor relationships on child mental health. Previous studies have come to similar conclusions: positive parent-student and teacher-student relationships often lead to signs of positive mental health in children. To expand upon these past findings, we used different measures of mental health and relationship qualities to create a predictive model. We conducted classification and coefficient weight analyses to see how strong of an impact different variables representing indicators of healthy relationships had on different aspects of child mental health. This information was used to create predictions of future cases. Like previous studies, we found that there was a general pattern that showed that good child-mentor relationships, defined specifically by the frequency of praise and fights, had an overall positive impact on child mental health, specifically when it came to symptoms of depression. Furthermore, parents seemed to have stronger impacts on child mental health than teachers did. Limitations include possibly biased respondents who may not be representative of the greater population, as well as the specificity of the variables that were chosen. Going forward, further steps to analyze different datasets and deepen the scope of the research will be helpful in finding patterns and developing more detailed conclusions.\n",
      "research_paper": "assets/pdfs/ae7c24c128051abc77e62973907114c6a2e90769.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://epe.brightspotcdn.com/59/83/84c4b26c46f7afb634762e6dd585/black-teacher-high-five-072023-1475870499.jpg",
      "headline": "In this paper, we use a dataset from the Substance Abuse and Mental Health Data Archive to analyze the impacts of child-mentor relationships on child mental health. Previous studies have come to similar conclusions: positive parent-student and teacher-student relationships often lead to signs of positive mental health in children. To expand upon these past findings, we used different measures of mental health and relationship qualities to create a predictive model. We conducted classification and coefficient weight analyses to see how strong of an impact different variables representing indicators of healthy relationships had on different aspects of child mental health. This information was used to create predictions of future cases. Like previous studies, we found that there was a general pattern that showed that good child-mentor relationships, defined specifically by the frequency of praise and fights, had an overall positive impact on child mental health, specifically when it came to symptoms of depression. Furthermore, parents seemed to have stronger impacts on child mental health than teachers did. Limitations include possibly biased respondents who may not be representative of the greater population, as well as the specificity of the variables that were chosen. Going forward, further steps to analyze different datasets and deepen the scope of the research will be helpful in finding patterns and developing more detailed conclusions.",
      "project_id": "05db0922f47faa75083fbb1384b9915d90f119a7",
      "expand": false,
      "tags": [
        "Mental Health",
        "Education",
        "Psych"
      ],
      "related_proj": [
        24,
        178,
        160
      ]
    },
    {
      "student_name": "Vincent P.",
      "mentor_name": "Amanda Wang",
      "mentor_title": "MS in Computer Science, Computer Science and Business Analytics MIT Alum",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Business and Finance",
        "Arts and Music"
      ],
      "project_title": "Analysis of Trending YouTube Videos: Finding Patterns in Viral Content",
      "project_desc": "As the digital world continues to grow, content creators frequently have trouble building a community and producing videos that will interest their audience.  Especially as these people look toward the internet for both recreational and monetary reasons, finding out techniques to build a community is important in today’s age. This paper analyzes the issues of video performance, revealing the patterns of what makes a video successful and viral. By training different models and testing different datasets, we were able to find the correlation between the potential chances of popularity and the video’s content. Using the most accurate model, the Random Forest model, content creators can see whether or not they are likely to do well based on patterns found in trending videos. ",
      "research_paper": "assets/pdfs/c283911c7c3a2f4a066583c76fef8f34767b6412.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.cnet.com/a/img/resize/55f2ab60e25d1ec7e18c4290c93e6e27300c8cc6/hub/2020/01/01/301595b2-19b5-4399-94a2-718b3e3ed15c/youtube-logo-laptop-4692.jpg?auto=webp&fit=crop&height=675&width=1200",
      "headline": "As the digital world continues to grow, content creators frequently have trouble building a community and producing videos that will interest their audience. Especially as these people look toward the internet for both recreational and monetary reasons, finding out techniques to build a community is important in today’s age. This paper analyzes the issues of video performance, revealing the patterns of what makes a video successful and viral. By training different models and testing different datasets, we were able to find the correlation between the potential chances of popularity and the video’s content. Using the most accurate model, the Random Forest model, content creators can see whether or not they are likely to do well based on patterns found in trending videos.",
      "project_id": "cc8726745726dab5f9cfc29ff8712b4823648242",
      "expand": false,
      "tags": [
        "Pop Culture",
        "Social Media",
        "Random Forests"
      ],
      "related_proj": [
        89,
        121,
        128
      ]
    },
    {
      "student_name": "Julia S.",
      "mentor_name": "Barbie Duckworth",
      "mentor_title": "Computer Science BS Alumni MIT, prior Software Engineer at Facebook and eBay",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Predicting NH4 Levels for Corn Crop in Wisc",
      "project_desc": "Ammonium (NH4), an organic matter that accumulates in the top portion of soil, can pose a serious risk to biodiversity. Using machine learning to construct regression models, NH4 levels can be predicted and therefore mitigated. In this paper we used linear, ridge, and lasso regressions. Through the evaluation of  crop farming factors that contribute to the NH4 levels, it was concluded that NO3 and N2O have the most direct correlation to NH4. These factors yielded the best accuracy for regression models with the best performing model being a multiple feature linear regression which resulted in 60% accuracy. While certain measures did improve the model’s performance, outliers continuously worsened the results.  \n \n",
      "research_paper": "assets/pdfs/9e9f02aa29e4f86d442c4c6c4e0a0968f454020d.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://thumbs.dreamstime.com/b/corn-field-drone-perspective-aerial-view-cultivated-green-landscape-140491930.jpg",
      "headline": "Ammonium (NH4), an organic matter that accumulates in the top portion of soil, can pose a serious risk to biodiversity. Using machine learning to construct regression models, NH4 levels can be predicted and therefore mitigated. In this paper we used linear, ridge, and lasso regressions. Through the evaluation of  crop farming factors that contribute to the NH4 levels, it was concluded that NO3 and N2O have the most direct correlation to NH4. These factors yielded the best accuracy for regression models with the best performing model being a multiple feature linear regression which resulted in 60% accuracy. While certain measures did improve the model’s performance, outliers continuously worsened the results.",
      "project_id": "78937f8c624c506b3879d74078159e88a7a6d2af",
      "expand": false,
      "tags": [
        "Earth Systems",
        "Farming",
        "Chemicals"
      ],
      "related_proj": [
        54,
        172,
        123
      ]
    },
    {
      "student_name": "Justin H.",
      "mentor_name": "Christina Cheng",
      "mentor_title": "Materials Science and Engineering PhD Candidate at Stanford, ",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Does Physiology Really Help a Volleyball Player Succeed?",
      "project_desc": "During my research project, I aimed to answer my main research problem of whether or not a person’s physiology affected their performance in women’s volleyball. It has always been assumed that a better physiology allows players to succeed in all sports, but I wanted to see if this was the case for volleyball. From previous research done by Sciendo, it has been shown that height is a core value for volleyball players, as it has a higher chance of helping a player succeed if they are taller. Although this is the case, they also found that Japanese volleyball players, who were shorter than the average professional volleyball player, were able to succeed and play at a high level. The overall approach used to answer this question was to see if there was any correlation between a player’s Ranking and their physiology (height, weight, spike, and block). I used a dataset named “Best Volleyball Players 2019” from Kaggle that displayed a wide range of women’s volleyball players and their different attributes along with their ranking. From the results that I have collected, the major conclusions in relation to my main question was that a person’s physiology did not affect their performance and ranking. The number for my error was large, meaning that you did not need to have a more optimal physiology to succeed in volleyball and be higher on the rankings. ",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://volleycountry.com/wp-content/uploads/2017/05/volleyball-attack-kurek-1080x650.jpg",
      "headline": "As a child, I always went to my sister’s volleyball games. From club games to school games, I went to them all. Because I saw so many volleyball games by watching my sister’s games, I always saw that there was always a trend that I kept seeing from each team.",
      "project_id": "0f5df92f9d5f434d7216389869ded3e6ce721194",
      "expand": false,
      "tags": [
        "Physiology",
        "Sports",
        "Vollyball"
      ],
      "related_proj": [
        67,
        59,
        7
      ]
    },
    {
      "student_name": "Saket R.",
      "mentor_name": "Bradley Yam",
      "mentor_title": "Computer Science BS from Yale, Software Engineer",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "What Factors Correlate with the Relationship Between Gender and Race and Pursuing STEM?",
      "project_desc": "The world of technology and Science, Technology, Engineering, and Math (STEM) is dominated by Caucasians, Asians, and males. According to Computer Science.org, while African Americans and Latinos make up 13.4% and 16% of the US population respectively, they only comprise 9% and 7% of STEM occupations (McGee). In this paper, I sought to answer the question, “How do races and genders differ in the way they pursue STEM, and what factors correlate with these differences?” Identifying which factors cause this lack of representation is the most important step in fixing the diversity problem. \n\nTo answer this question, I used data from the High School Longitudinal Study (HSLS), which asked hundreds of questions at different points throughout students’ high school and college careers. I used 16 of these questions and analyzed how strongly they correlate with pursuing a STEM career or not pursuing a STEM career. I found that Asians are much more likely to pursue STEM, African Americans are much less likely to pursue STEM, and women are much more likely to pursue STEM, although they largely pursued health occupations within STEM.\n\nIt is important that we make sure everyone feels comfortable pursuing or not pursuing STEM. From my literature review, I found that Asians are more likely to and African Americans are less likely to pursue STEM because Asians feel like they fit in and African Americans feel like they don’t (Milgrom-Elcott). In order to increase the diversity in the STEM workforce, we need to make sure all minority groups feel welcomed in high school STEM classes, college STEM courses, and STEM workplaces themselves.\n",
      "research_paper": "assets/pdfs/5b2b8e16f39c1db60406f1020de89a4abb9cf94d.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.familyeducation.com/sites/default/files/2023-02/Black%20Female%20Scientists%20That%20Have%20Shaped%20US%20History_Feature.jpg",
      "headline": "In this paper, I sought to answer the question, “How do races and genders differ in the way they pursue STEM, and what factors correlate with these differences?” Identifying which factors cause this lack of representation is the most important step in fixing the diversity problem. ",
      "project_id": "6a81620dda02871bb7ce4f1b2c806a64ac08f1be",
      "expand": true,
      "tags": [
        "Gender Equality",
        "Racial Equality",
        "Education",
        "Social Justice"
      ],
      "related_proj": [
        35,
        133,
        126
      ]
    },
    {
      "student_name": "Peijie G.",
      "mentor_name": "Kush Khosla",
      "mentor_title": "MS Computer Science from Stanford, Lead Data Scientist at Retain.ai",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": "Predicting Future Phonological Changes Of Mandarin Chinese ",
      "project_desc": "Linguists made historical reconstructions extensively based on existing patterns and common trends of phonological changes without the aid of sound data, yet previously little was done to extend such patterns into the future as a result of mostly human work in the field of historical linguistics. We created a system of translating IPA representation to vectors that captures characteristics of phonemes such as their articulatory location, and experimented with several machine learning models to capture existing trends from Old, Middle to Modern Chinese (Mandarin). Then we attempted to predict future Mandarin pronunciations based on existing phonological data and ancient reconstructions. The models were successful at capturing trends of Ancient to Modern Chinese shifts, as a great majority of the validation results were exactly correct. The predictions from the models vary in consistency of outputs and phonological inventory, but they are largely logical and may offer insight into future shifts. Therefore it is possible to predict phonological changes with machine learning, yet the accuracy of the predictions still needs validation in the future. If the predictions are proven to be effective in the future, the solutions may provide a method of capturing phonological changes and validating previous ancient reconstructions. ",
      "research_paper": "assets/pdfs/91ca19e3cc5325d7c9ee52573093be4174269b43.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://sguru.org/wp-content/uploads/2019/11/mandarin.jpg",
      "headline": "We created a system of translating IPA representation to vectors that captures characteristics of phonemes such as their articulatory location, and experimented with several machine learning models to capture existing trends from Old, Middle to Modern Chinese (Mandarin). ",
      "project_id": "01032d50e1636761d051f258a4ed525c6834fac1",
      "expand": false,
      "tags": [
        "Natural Language Processing",
        "Modern Languages"
      ],
      "related_proj": [
        84,
        51,
        109
      ]
    },
    {
      "student_name": "Archith S.",
      "mentor_name": "Barbie Duckworth",
      "mentor_title": "Computer Science BS Alumni MIT, prior Software Engineer at Facebook and eBay",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/4744640f73ce5f3d6f86765dad1d4f1604b083a9.png",
      "domains": [
        "Environment",
        "Healthcare and Biology"
      ],
      "project_title": "Insect Identification Project for Agricultural Advancement",
      "project_desc": "Insects are complex creatures whose influence and characteristics can drive ecosystems or destroy plant populations. Therefore, it is imperative to recognize the differences between them and whether they pose harm or benefits, yet the majority of humans do not have a keen view for the differences in insect qualities. Through the use of image data, we developed an artificial intelligence system which is able to predict an insect’s species based on a photo of a given insect. To assess the image data, we developed a convolutional neural network (CNN) which uses a series of neurons to output a result which we can interpret to identify the given insect. The highest validation accuracy we received was approximately 80%, which given the very subtle differences between the insects we used, the amount of categories we were defining from, and the limited number of images in each category, was a decently high validation accuracy. From this, we could conclude that we can not only identify a species, but with relatively high accuracy, giving us insight into how insect identification is a valid and applicable process which can be implemented into agricultural and pest control technology. ",
      "research_paper": "assets/pdfs/b1f9b06b9b68ef844fbc4a55a3f282e47ef9af0a.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/archith-s/Insect-Identification",
      "graphic_link": "https://www.innovationnewsnetwork.com/wp-content/uploads/2022/02/beekeeping.jpg",
      "headline": "Through the use of image data, we developed an artificial intelligence system which is able to predict an insect’s species based on a photo of a given insect. ",
      "project_id": "e74fd4dd5f1bdd0250ad1c870497b02257280bb2",
      "expand": false,
      "tags": [
        "Wildlife",
        "Farming"
      ],
      "related_proj": [
        48,
        183,
        42
      ]
    },
    {
      "student_name": "Eera B.",
      "mentor_name": "Clayton Greenberg",
      "mentor_title": "Teaching Professor from UPenn",
      "mentor_image": "assets/images/mentor_imgs/c41d5535b4ef0054a8ee46918a196c8732eb5608.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Environment"
      ],
      "project_title": "Plant Toxicity Classification by Image",
      "project_desc": "Poisonous and non-poisonous plants have extremely similar visible features to any non-botanist,\nwhich puts those in danger who are frequently present in areas containing various plants. Failure\nto distinguish harmful plants from safe ones puts several people at a high risk of accidents and\npotential health issues after contact with a toxic plant. While previous work has found ways to\nclassify specific types of plants, a limited amount of research has been done with toxic and\nnon-toxic plants of several species. Since differentiating between dangerous and safe plants is a\ncomplex task for a human brain, this study approaches the issue through machine learning\nmodels starting with a convolutional neural network (CNN) and discovering that a logistic\nregression model—trained on a dataset with manually designed features—has the best\nperformance with the particular dataset used. The neural network demonstrated overfitting which\nwas likely caused by the inconsistent backgrounds of images within the dataset. The logistic\nregression model achieved an accuracy of 97.37% in its predicted classifications for the plants.\nThe best-performing logistic regression model contained the three-leaf and dark red stem\nfeatures indicating that these two features are the most reliable ones used to distinguish between\ntoxic and non-toxic plants.",
      "research_paper": "assets/pdfs/0c1290eefaf9ec7f2aef9d2635e0936e206d83dc.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://post.healthline.com/wp-content/uploads/2022/10/angel-trumpet-plant-flowers-1200-628-facebook-1200x628.jpg",
      "headline": "Since differentiating between dangerous and safe plants is a complex task for a human brain, this study approaches the issue through machine learning models starting with a convolutional neural network (CNN) and discovering that a logistic regression model—trained on a dataset with manually designed features—has the best performance with the particular dataset used. ",
      "project_id": "71e4256690404e931a1ba7dbc8524aa7f15eb54b",
      "expand": false,
      "tags": [
        "Earth Systems",
        "Plants",
        "Classification"
      ],
      "related_proj": [
        160,
        17,
        101
      ]
    },
    {
      "student_name": "Eve B.",
      "mentor_name": "Tomer Arnon",
      "mentor_title": "MS Engineering at Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/36fa5be484ece500f31fbf51c6d6ff307ad7ba06.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Using Machine Learning to identify the Habitability of Exoplanets from TESS Transit Data",
      "project_desc": "With so many new tools and methods to detect exoplanets, it is time to start determining which are deserving of more research in the hunt for other habitable planets. We began our research with exoplanets that are similar to Earth as the most likely to house life. We used transit measurements such as an exoplanet's size, temperature, and relative distance from its star as well as machine learning to develop two models: one to predict if an object of interest was an exoplanet and another to cluster our data to determine its similarity to Earth. We achieved an 88.7% accuracy for the first model, and our clustering algorithm found interesting results when compared to a graph showing the principal components of the data. We found that our Earth data point fell right on the origin of our 2 dimensional PCA graph and was placed in our largest cluster. Both of our models showed that machine learning can be useful in the search for habitable exoplanets and is deserving of more research.",
      "research_paper": "assets/pdfs/8f548c4b9bf6ae3e084359150b1c38b47eca2e28.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://scitechdaily.com/images/How-Scientists-Search-for-Habitable-Planets.jpg",
      "headline": "With so many new tools and methods to detect exoplanets, it is time to start determining which are deserving of more research in the hunt for other habitable planets.",
      "project_id": "f8e34d726f8105d43be6e3ca5c92e7ff69d4dc6c",
      "expand": true,
      "tags": [
        "Astronomy",
        "Physics"
      ],
      "related_proj": [
        20,
        166,
        91
      ]
    },
    {
      "student_name": "Maximilian P.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Predicting Solar Array Output Using Weather Sensor Data",
      "project_desc": "With the recent push for renewable energy sources, solar energy is one that is readily available. This paper will explore how to predict the power output of a solar array based on weather data, collected from sensors throughout each day.. With the predicted output of a solar array, a grid operator could decide which other sources of energy would be needed to run to meet energy demand. A linear regression machine learning model will be used to create a correlation between weather data such as irradiation and power output. The best model, the Gradient Boosting Regressor, produced a mean average error of just under 300w.",
      "research_paper": "assets/pdfs/6008428e114c5261e67c9729cab14272a1574e37.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.sunrun.com/sites/default/files/styles/gsc_feature_750x300/public/best-solar-panels-home-budget-feature-opt.jpg?itok=GR6g7v48",
      "headline": "With the recent push for renewable energy sources, solar energy is one that is readily available. This paper will explore how to predict the power output of a solar array based on weather data, collected from sensors throughout each day.",
      "project_id": "028983fdbb7bceb940cab8c91909f93a229c83af",
      "expand": false,
      "tags": [
        "Solar Cycles",
        "Solar Power",
        "Solar Flares",
        "Earth Systems"
      ],
      "related_proj": [
        37,
        172,
        111
      ]
    },
    {
      "student_name": "Claire M.",
      "mentor_name": "Clayton Greenberg",
      "mentor_title": "Teaching Professor from UPenn",
      "mentor_image": "assets/images/mentor_imgs/c41d5535b4ef0054a8ee46918a196c8732eb5608.png",
      "student_image": "assets/images/student_imgs/56eba6fde8c67ede41242e34e609b24ce0f4bfad.png",
      "domains": [
        "Psychology"
      ],
      "project_title": "Predicting emotion ratings from  color statistics of images",
      "project_desc": "Which features in art affect emotional reactions? In order to answer this question, we took a variety of abstract images and included people’s emotional ratings for each image, for eight different emotions. We utilized this data to see how visual elements affect our everyday lives by engineering features of images and running that through a variety of neural networks. Out of all the networks we ran our data through, the linear regression model performed the best with a mean average error of around 1.4. All the other neural networks, including a convolutional neural network, had a mean average error of at least 1.7. The full-size linear regression model with all engineered features worked best in predicting the emotional values associated with each image. That is, all specific features were significant and we did not find meaningful differences among the models for different emotion ratings.",
      "research_paper": "assets/pdfs/84051a62e1d5babdd5177b3383fbb7dbd5712c8f.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.searchenginejournal.com/wp-content/uploads/2019/07/google-to-capture-and-learn-about-our-emotions-on-a-smartphone-camera.png",
      "headline": "We utilized this data to see how visual elements affect our everyday lives by engineering features of images and running that through a variety of neural networks.",
      "project_id": "a9e8f0362783db914e4fc941e2d7e176e13f0e8a",
      "expand": false,
      "tags": [
        "Psych",
        "Facial Emotion"
      ],
      "related_proj": [
        103,
        84,
        127
      ]
    },
    {
      "student_name": "Thalia S.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "PoliSci and Law"
      ],
      "project_title": "Is AI Necessary In Deciding Whether an Offender Is Likely To Recidivate, With & Without the Effect of Protected Characteristics",
      "project_desc": "Our research topic addresses the issue of whether or not an AI algorithm can be developed  to forecast whether or not a criminal would recidivate by removing protected characteristics from models in terms of both accuracy, precision and recall and their equality across different groups, as well as if such a tool is necessary. The background of this issue is that there is always an underlying human prejudice present in the criminal justice system when it comes to making judgments, thus it is crucial to determine whether or not this human bias is required while making these decisions and whether there is similar kind of bias with machine learning methods. There are various ways to approach and respond to this topic, but the strategy we used involved examining the key variables that might influence prejudice, such as gender or ethnicity, and comparing how well the AI system can predict whether someone is likely to recidivate with or without incorporating these protected characteristics. The research's significant findings are that, after utilizing a number of different models, logistic regression was used to determine whether or not an offender was likely to recidivate. The accuracy of the predictions for our logistic regression model was 67.74 percent without the removal of any protected characteristics that hold risk of being discriminated against unfairly, but it increased to 68.16 percent when race and gender were taken into account. The random Forest model revealed that, after accounting for the gender characteristic, the mean absolute error—the percentage of variance between each measurement and the quantity's actual value—was 34.06 percent along with an accuracy score of 65.94 percent. The main results drawn from this demonstrate that utilizing various models and measuring the precision, recall, and accuracy of each race and gender are comparable with or without protected characteristics. Although the accuracy of this still isn't the finest, it could demonstrate that bias may not be as significant a factor in decision-making as it may first appear. This, however, can also be contested because the dataset utilized to train these models could be encoding real world bias such that race and gender get encoded into other variables such as prior offenses; this would mean that even if we removed race and gender data it would still be present in our model through other variables. The question of whether or not we should remove protected characteristics is required and needed remains unanswered as a result.",
      "research_paper": "assets/pdfs/331be37a971db7be1b735d8c9e28933823b48c02.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.forbes.com/advisor/wp-content/uploads/2022/12/Criminal_Justice.jpeg-1.jpg",
      "headline": "Our research topic addresses the issue of whether or not an AI algorithm can be developed  to forecast whether or not a criminal would recidivate by removing protected characteristics from models in terms of both accuracy, precision and recall and their equality across different groups, as well as if such a tool is necessary",
      "project_id": "1c112a8fe4e6a3353c722d06b6f53e75d25bc422",
      "expand": true,
      "tags": [
        "Criminal Justice",
        "Recidivism",
        "Social Justice"
      ],
      "related_proj": [
        114,
        100,
        78
      ]
    },
    {
      "student_name": "Noor E.",
      "mentor_name": "Yiran Li",
      "mentor_title": "Biometry and Stats Cornell Alum, Bioinformatics MS at ETH Zürich",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/958b5b973722fdda18a24049d1d011e96faf6856.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Application of Machine Learning to Distinguish Premature Leukemia Cells from Healthy Blood Cells",
      "project_desc": "Leukemia is a disease that primarily impacts children. The likelihood of beating this deadly disease increases with early detection. Because premature Leukemia blasts are unidentifiable by the human eye, different methods are needed to accurately diagnose cancer in its early stages. Convolutional Neural Networks (CNNS) to solve this problem have been created with classification validation accuracy rates as high as 96.15%. A dataset from the Cancer Imaging Archive [1] was used to train a CNN model. Of the 15,135 images in the dataset, 600 were used to prevent system crashing. With the limited resources available, a 63.6% validation accuracy was achieved. ",
      "research_paper": "assets/pdfs/abf74209725c307c0eb1d3fc2b5744f759e11777.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://cdn2.poz.com/40679_iStock-117868474.jpg_4145d49a-c89c-40eb-a574-290f096de470_x2.jpeg",
      "headline": "Convolutional Neural Networks (CNNS) to solve this problem have been created with classification validation accuracy rates as high as 96.15%. ",
      "project_id": "7a409f003cf56c704a89210b564fa7ef5ddec77e",
      "expand": true,
      "tags": [
        "Leukemia",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        97,
        137,
        24
      ]
    },
    {
      "student_name": "Neal D.",
      "mentor_name": "Clayton Greenberg",
      "mentor_title": "Teaching Professor from UPenn",
      "mentor_image": "assets/images/mentor_imgs/c41d5535b4ef0054a8ee46918a196c8732eb5608.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Cybersecurity"
      ],
      "project_title": "Identifying EOL software",
      "project_desc": "Throughout our project, we have been exploring ways to identify EOL and malicious websites using python. With this software, we can help people stay away from these types of websites. There are lots of cybersecurity problems in the world, but we wanted to address the EOL problem because it is something that is within the realm of what we are learning (language recognition), and it has a practical use in the world. We used multiple different commands to identify certain “keywords,” with the intention of getting the highest possible accuracy percentage. We also prioritized the elimination of false negatives over false positives. Creating a new variable with the keywords, we were able to correctly predict 99% of the EOL websites using the data given to us. We were able to conclude that mixing the 2 most important data columns and creating one variable to determine both of the variables is the best way to go about creating a variable that gives you the most.",
      "research_paper": "assets/pdfs/d34a47e0378e43f257434214043f08ba4ac4aaee.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://auxzillium.com/wp-content/uploads/2019/07/eol-image-WINDOW-7-1030x504.jpg",
      "headline": "Throughout our project, we have been exploring ways to identify EOL and malicious websites using python. Creating a new variable with the keywords, we were able to correctly predict 99% of the EOL websites using the data given to us.",
      "project_id": "2e33ac90723516d3347d5bd42fabc3f4163edd87",
      "expand": false,
      "tags": [
        "Natural Language Processing",
        "EOL",
        "Web Development"
      ],
      "related_proj": [
        165,
        173,
        139
      ]
    },
    {
      "student_name": "Deniz G.",
      "mentor_name": "Kush Khosla",
      "mentor_title": "MS Computer Science from Stanford, Lead Data Scientist at Retain.ai",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Education",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "The Impact of the Covid-19 Pandemic on the Test Scores of Various Demographics",
      "project_desc": "The research problem that this research paper attempts to address is, How did Covid impact and affect the test scores of different demographics of students in various schools in New York? This issue is significant for the following reasons. By the conclusion of the school year, students had an average of five months’ worth of math and four month’s worth of reading behind them, according to an analysis of the pandemic’s effects on K-12 student learning. For example, student testing in 2021 was about ten points behind in math and nine points behind in reading, compared with matched students in previous years. The pandemic expanded already-existing achievement and opportunity inequities, particularly harming historically underprivileged students. Children in low-income schools had seven months of unfinished arithmetic work at the end of the year, compared to six months for students in majority-Black schools. High school students are now more likely to drop out, and seniors, particularly those from low-income households, are less likely to continue their education after graduation. The crisis also had an effect on pupils’ general health and well-being, with more than 35% of parents expressing severe or great concern for their kids’ mental health. The effects of the epidemic pose a threat to the hopes and possibilities for this generation far into adulthood. Their chances of enrolling in college and subsequently securing a rewarding profession that allows them to sustain a family may be adversely affected by the knock-on effects. Furthermore, analysis indicates that the pandemic’s effects on schooling may result in today’s students earning $49,000 to $61,000 less in their lifetime unless measures are implemented to remedy unfinished learning. The overall approach that was used to answer this question involved using statistical analysis to examine the correlation between how certain factors such as being economically disadvantaged or being hispanic affected student proficiency scores on New York State regents exam tests in 2018 and 2021 in different schools. Some of the significant results that came out of this research paper involved certain demographics such as those who were economically disadvantaged and those who were white. Specifically, in relation to how their demographic correlated with their school’s percent proficiency difference in a subject exam between the years 2018 and 2021. For example, 100% of the students of the Academy for Language and Technology are economically disadvantaged and compared to 2021, the number of students who were proficient in English dropped by 49%. Furthermore, 72% of Akron High school’s students are economically disadvantaged and compared to 2021, the number of students who were proficient in English dropped by 68%. However, on the other side of the spectrum, students of white descent did not fare as badly. For instance, Adirondack Middle School has an approximate demographic makeup of 100% white students. Their student’s percent proficiency when it came to their Algebra I score did not change at all. Additionally, Yonkers Montessori Academy has a demographic makeup where 46% of its students are white. The students from this school showed a 5% increase in their proficiencies in their science scores. The research that was conducted as the basis for this paper was done in an effort to examine the COVID-19 pandemic’s effects on student performance across various schools in New York. Specifically, it affected some demographics disproportionately when compared to others. Due to students being forced to learn from home, some without the necessary technology and adequate resources. Consequently, a lot of students took a significant hit to their academic progress. For instance, 100% of a student’s demographic make up is economically disadvantaged and out of these students, 49% of them were no longer proficient in ELA after the COVID-19 pandemic. Whereas, students of white descent did not fare as badly.  For instance, Adirondack Middle School has an approximate demographic makeup of 100% white students. Their student’s percent proficiency when it came to their Algebra I score did not change at all.",
      "research_paper": "assets/pdfs/e8d80e23bb221e30b11b74806ba7ab11dcd7ac0a.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcTu1QsK4minVxl1ddZ7leeN_hYkVq4-EHb6QFg1A1_gpQ&usqp=CAU&ec=48665701",
      "headline": "The research problem that this research paper attempts to address is, How did Covid impact and affect the test scores of different demographics of students in various schools in New York? The overall approach that was used to answer this question involved using statistical analysis to examine the correlation between how certain factors such as being economically disadvantaged or being hispanic affected student proficiency scores on New York State regents exam tests in 2018 and 2021 in different schools.",
      "project_id": "cf9350cead2fe96bbfaf975fb68dd1f94d126b3e",
      "expand": true,
      "tags": [
        "Education",
        "COVID-19",
        "Racial Equality",
        "Inequality",
        "Wage Gap",
        "Gender Equality",
        "New York"
      ],
      "related_proj": [
        133,
        40,
        65
      ]
    },
    {
      "student_name": "Serena F.",
      "mentor_name": "Clayton Greenberg",
      "mentor_title": "Teaching Professor from UPenn",
      "mentor_image": "assets/images/mentor_imgs/c41d5535b4ef0054a8ee46918a196c8732eb5608.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Arts and Music",
        "Engineering"
      ],
      "project_title": "Brightness helps CNN classify a subset of the images from Google Quick Draw",
      "project_desc": "We made a CNN that learned to recognize and classify sketches from Google’s Quick Draw dataset and implemented a novel brightness feature to test if it increased the accuracy. The results displayed up to a 4% increase in validation accuracy. To conduct experiments, we created 3 models, 1 with brightness and two without. In addition, we used different numbers of epochs, metalayers, kernels, and sizes of kernels and dropouts. ",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.labnol.org/images/2023/google-drive-drawings.jpg",
      "headline": "We made a CNN that learned to recognize and classify sketches from Google’s Quick Draw dataset and implemented a novel brightness feature to test if it increased the accuracy.",
      "project_id": "eb405932a81ebf0ee021c4d7110eb79a8be94072",
      "expand": false,
      "tags": [
        "Google",
        "Visual Arts",
        "CNNs",
        "Classification"
      ],
      "related_proj": [
        161,
        62,
        159
      ]
    },
    {
      "student_name": "Emily J.",
      "mentor_name": "Philip Bell",
      "mentor_title": "Computer Science MS from Georgia Tech, Harvard and Oxford alum, AI Engineer at Cognino ",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/b840b9aa2dbebfd0840ee3d29796d5d0522107f0.png",
      "domains": [
        "Education"
      ],
      "project_title": "Texas High School Dropout Rates",
      "project_desc": "Dropout rates in high schools throughout Texas have been steadily increasing, especially following the Covid-19 pandemic. Dropping out of high school has a permanent effect on the future of a person’s life, and as more students don’t make it to graduation, the more serious this issue becomes. The only way to combat this problem is by identifying what factors or characteristics of a campus are causing students to drop out. This project aims to solve this problem through an artificial intelligence algorithm that can predict the dropout rate of a Texas high school campus based on specific characteristics of the school. This project uses a gradient boosting model on a dataset by the TEA (Texas Education Agency) with information about high schools throughout the state. The results of this model are also implemented into an app where people can input information about a school and get a dropout rate prediction, as well as recommendation for that specific campus on how to decrease dropout rates. ",
      "research_paper": "assets/pdfs/87879ba876a94ace099efd8b0389d59c655d6fd5.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/emily-joseph/Texas_Dropout_Rates/blob/main/app.py",
      "project_webpage": "https://emily-joseph-texas-dropout-rates-app-l52cuq.streamlit.app/",
      "graphic_link": "https://newjerseymonitor.com/wp-content/uploads/2021/12/GettyImages-168943719-700x467-1.jpg",
      "headline": "Dropout rates in high schools throughout Texas have been steadily increasing, especially following the Covid-19 pandemic. This project aims to solve this problem through an artificial intelligence algorithm that can predict the dropout rate of a Texas high school campus based on specific characteristics of the school.",
      "project_id": "4cc59fab0643e8de76419946644772a98630fb3b",
      "expand": false,
      "tags": [
        "Education",
        "COVID-19",
        "Dropout Rates"
      ],
      "related_proj": [
        40,
        75,
        159
      ]
    },
    {
      "student_name": "Yudhiishbala V.",
      "mentor_name": "Ana Sofia Muñoz Valadez",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/dc572bb7e79beebf858669c93a28204288f6e88e.png",
      "domains": [
        "PoliSci and Law"
      ],
      "project_title": "AI Models Necessary to Reduce Racism in Criminal Justice In The United States of America",
      "project_desc": "As this project commenced, I first focused on gaining some general knowledge about the applications of Artificial Intelligence. As I went through this, I ended up choosing The main focus of this project was to figure out which kind of AI model is best for use in a court setting or the criminal justice system.The goal was to ensure the AI model is accurate and free of skewed/biased data. Throughout this research project, three different regression models are looked at in order to narrow down the most accurate models. This research found that the AI models that can provide the most accurate results are DecisionTreeRegressor, MLPRegressor, and BaggingRegressor. As this research project commenced, it was shown that the BaggingRegressor method enables programmers to train artificial intelligence systems to make decisions in the most accurate manner possible, with the margin of errors and biases decreasing by almost 82% from the MLPRegressor.",
      "research_paper": "assets/pdfs/0c25a65e391d1ca6d60de881e00e9b37dd37a931.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://i0.wp.com/www.goodwin.edu/enews/wp-content/uploads/2022/03/Benefits-of-a-Criminal-Justice-Degree_mast-tiny.jpg?fit=1920%2C1280&ssl=1",
      "headline": "The goal was to ensure the AI model is accurate and free of skewed/biased data. Throughout this research project, three different regression models are looked at in order to narrow down the most accurate models.",
      "project_id": "c074529d01f1911b9dead55c4b7171d4168250c2",
      "expand": true,
      "tags": [
        "Criminal Justice",
        "Law and Order",
        "Racial Equality"
      ],
      "related_proj": [
        47,
        72,
        79
      ]
    },
    {
      "student_name": "Aman B.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/student_imgs/1fb3f8849768af5f249a603848573804362d8d41.png",
      "domains": [
        "PoliSci and Law",
        "Language and Literature"
      ],
      "project_title": "Building an Optimized algorithm that provides summaries of legal documents",
      "project_desc": "We analyze the accuracy of various NLP algorithms in providing text summarization and fine-tune a particular model on a dataset to provide accurate text summaries of legal documents. The legal industry is built around documents as they provide evidence and reduce doubt in the court. Due to the large volume of documentation in the legal industry, the processing and summarization of these documents is important to a number of individuals. For example, lawyers, clients and professionals may need access to summaries of the documents for reference to similar cases. In this paper, we have developed an algorithm that trains the T5 model on the legal domain in order to create more accurate summaries of legal documents. We were able to create a user interface that allows for the input of documents and makes use of the algorithm we created to output a summary of the document which can be copied by the user.",
      "research_paper": "assets/pdfs/3eec95bc01e8a7ab3ad6093aa2ffb69309459447.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/aman8533/Building-an-Optimized-algorithm-that-provides-summaries-of-legal-documents",
      "graphic_link": "https://scitechdaily.com/images/Legal-Document-Concept.jpg",
      "headline": "The legal industry is built around documents as they provide evidence and reduce doubt in the court. Due to the large volume of documentation in the legal industry, the processing and summarization of these documents is important to a number of individuals. We were able to create a user interface that allows for the input of documents and makes use of the algorithm we created to output a summary of the document which can be copied by the user.",
      "project_id": "de8eddad7c2dad6d67decd1f64ce8124032c2129",
      "expand": false,
      "tags": [
        "Law and Order",
        "Natural Language Processing"
      ],
      "related_proj": [
        162,
        66,
        100
      ]
    },
    {
      "student_name": "Viksar D.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/student_imgs/55f9c79407f4d88446c84968a146990e9ff152ff.png",
      "domains": [
        "Education"
      ],
      "project_title": "Predicting Dropouts Using Machine Learning Models",
      "project_desc": "High dropout rates in high schools and universities have become a major complication in many countries following the upsurge of the Covid 19 pandemic. A student’s experience at school is one of, if not, the largest contributors to the likelihood that they drop out. However, with the growing quantity of students in school, and the sheer variety of those who drop out, manually pinpointing possible dropouts becomes extremely challenging. Therefore, by using a machine learning model, administrators would be able to more efficiently identify a possible dropout and provide the necessary resources to ensure their success. The data, acquired from the UCI Machine Learning Depository, had 24 inputs, and 3630 samples. For this particular project, Logistic Regression, K-Nearest Neighbors, Naive Bayes, Extra Trees Classifier, and an MLP Neural Network models were tested. Among these models, the most accurate was the Extra Trees Classifier model, with a percentage accuracy of 77.41%, a loss of 7.85, and an F1 score of 82.35%. With the accuracy provided, this model should be held in high regard as it serves as an efficient method for predicting possible college dropouts in university.",
      "research_paper": "assets/pdfs/b545fd268b01bf9a88c99db2620adce4ac66f5b0.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "project_webpage": "http://6bad-34-91-124-160.ngrok.io/",
      "graphic_link": "https://static01.nyt.com/images/2018/05/25/opinion/25leonhardt-newsletter/merlin_35629513_5bc654be-2bee-42b7-b203-8a443cd862eb-superJumbo.jpg",
      "headline": "High dropout rates in high schools and universities have become a major complication in many countries following the upsurge of the Covid 19 pandemic. With the accuracy provided, this model should be held in high regard as it serves as an efficient method for predicting possible college dropouts in university.",
      "project_id": "d390a53d3c26a270b7cd24698fc9696b884b86ea",
      "expand": false,
      "tags": [
        "Education",
        "COVID-19",
        "Dropout Rates",
        "Logistic Regression"
      ],
      "related_proj": [
        159,
        86,
        40
      ]
    },
    {
      "student_name": "Roshni K.",
      "mentor_name": "Philip Bell",
      "mentor_title": "Computer Science MS from Georgia Tech, Harvard and Oxford alum, AI Engineer at Cognino ",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Journalism",
        "Language and Literature"
      ],
      "project_title": "Fake News Classification",
      "project_desc": "As fake news becomes more of a problem across the US, its consequences are becoming more and more damaging. Many viewers who interact with information online are unable to distinguish the difference between fake information and real information, especially with social media worsening the spread of fake articles. This project aims to combat this disinformation through an artificial intelligence algorithm that can classify real articles versus fake ones. The model utilizes a BERT model and is tested on two different datasets, the Kaggle dataset and the LIAR dataset. The results are also presented through an app where readers can input any article and immediately find out whether it contains false information or not.",
      "research_paper": "assets/pdfs/3979a93a7065e34965988a5f02311bf3708b6051.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://media.newyorker.com/photos/59097c441c7a8e33fb390384/master/w_4954,h_2560,c_limit/FakeNews_Lemann_WP.jpg",
      "headline": "As fake news becomes more of a problem across the US, its consequences are becoming more and more damaging. This project aims to combat this disinformation through an artificial intelligence algorithm that can classify real articles versus fake ones.",
      "project_id": "85fb41c71beb68825639358f04f6b3dce97a4b31",
      "expand": false,
      "tags": [
        "News",
        "Fake News",
        "Natural Language Processing",
        "LSTMs"
      ],
      "related_proj": [
        57,
        87,
        109
      ]
    },
    {
      "student_name": "Christopher G.",
      "mentor_name": "Sean Konz",
      "mentor_title": "Computer Science MS from Stanford, Software Engineer",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/bb598a63aa0d8f96d36cdff884f20beeed0418a8.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": "Generation of Research Paper Titles",
      "project_desc": "Can NLP accurately and effectively generate research paper titles? In this research paper, an effective and accurate artificial intelligence NLP model is tried to be determined by evaluating various models and methods for title generation. Titles are a vital part of any work. People tend to struggle to find good titles for their work, including researchers. A title must be representative of the whole paper and its topic, so it must be carefully chosen. Titles are the main reason people read a paper, so it’s very important to attract the audience of the researcher. Apart from this, normalizing research paper titles with generated ones can eliminate bias and exaggeration from them. To determine an accurate and effective model, various methods for title generation were tested given the abstract of a research paper. The methods tested vary from very simple, like “bag of words”, to complex with GPT-2. The best result was getting a sentence similarity of 0.9032, which was 66% higher than the initial model. Through this project it is confirmed that generating good titles with NLP is a hard task, but the results from the models can facilitate the process of creating a good and compact research paper title.",
      "research_paper": "assets/pdfs/71074e8b5e6842508b82111d5b48adabaeca6d08.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/cgossmann05/research-paper-title-generation",
      "graphic_link": "https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQD4fPz0UVShi0FJ6j8_ApzNM99E6LGl9m9Wn-lyjbalw&usqp=CAU&ec=48665701",
      "headline": "Can NLP accurately and effectively generate research paper titles? In this research paper, an effective and accurate artificial intelligence NLP model is tried to be determined by evaluating various models and methods for title generation. ",
      "project_id": "de4b0165aff73fddc32d35a35376d1647cd002f8",
      "expand": false,
      "tags": [
        "Natural Language Processing",
        "Generative AI",
        "GPT-3"
      ],
      "related_proj": [
        162,
        79,
        38
      ]
    },
    {
      "student_name": "Chinmay R.",
      "mentor_name": "Amanda Wang",
      "mentor_title": "MS in Computer Science, Computer Science and Business Analytics MIT Alum",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Using Machine Learning to Classify  Stars, Quasars, and Galaxies ",
      "project_desc": "In this project, we looked at data from stars, quasars, and galaxies from the sixteenth data release of the Sloan Digital Sky Survey Telescope. The aim of the project was to accurately and quickly classify these three types of objects using machine learning. We used three machine learning algorithms, namely logistic regression, multi-layer perceptron, and  decision tree classifier. The classification was done based on measurements of the object’s redshift and its u,g,r,i, and z light emissions. Logistic regression offered the least accurate results, with an overall accuracy of 94.83%, and a runtime of 28.2 seconds. Better in both aspects was the decision tree classifier, with an overall accuracy of 98.90%, and a runtime of 6.36 seconds. Several structures with different kinds of neuron-layer arrangements were tried for the MLP classifier. While all yielded around a 98% total accuracy, the more complex a structure was, or the more nodes and layers it had, the more time it took to run, and none even came close to the low amount of time taken by the decision tree classifier.\n",
      "research_paper": "assets/pdfs/8d1a6c6dcdfa827dfa8e15714053c5ea70b13af9.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://imageio.forbes.com/specials-images/imageserve/6448c936e1113d7182a3babe/Artist-s-rendering-of-quasar-P172-18--credit--ESO-M-Kornmesser--higher-res--1-/960x0.jpg?format=jpg&width=960",
      "headline": "In this project, we looked at data from stars, quasars, and galaxies from the sixteenth data release of the Sloan Digital Sky Survey Telescope. The aim of the project was to accurately and quickly classify these three types of objects using machine learning. We used three machine learning algorithms, namely logistic regression, multi-layer perceptron, and  decision tree classifier. ",
      "project_id": "88fdc975a0663bc34d05cc15d62b26f5e1807526",
      "expand": false,
      "tags": [
        "Astronomy",
        "Stellar Classification",
        "Logistic Regression",
        "Classification"
      ],
      "related_proj": [
        20,
        53,
        1
      ]
    },
    {
      "student_name": "Parinita K.",
      "mentor_name": "Philip Bell",
      "mentor_title": "Computer Science MS from Georgia Tech, Harvard and Oxford alum, AI Engineer at Cognino ",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Psychology",
        "Language and Literature"
      ],
      "project_title": "Can A Person’s MBTI Type Be Determined By A Sample Of Their Writing?",
      "project_desc": "Personality typology, in this case MBTI, can give us great insight into how a person thinks and operates and what motivates that person’s decisions.  Although typology is most commonly seen in pop psychology and used for fun, typing backed by more empirical evidence and less by self-reported answers, which rely solely on the self-awareness and unbiasedness of the person, could give a more accurate insight into how people fit into the theory of cognitive functions, which MBTI is based on.  This project aims to lessen the reliance of self-report for personality typology through an artificial intelligence algorithm that can type people as one of the 16 MBTI types using an unedited writing sample by that person. The model used in this project is a transformer model, and the dataset used is a Kaggle dataset consisting of 8000 people’s personality types and 50 written samples from each person. The results are presented through a Streamlit app that uses the model to assign a personality type according to inputted text.  The accuracy measured while testing this model is approximately 57.64%, which means the model correctly classified over half the text samples it was tested on.  This indicates that there is a pattern that the model is picking up on, other than complete chance, that connects a person’s writing to their MBTI type. However, the writing is not 100% predictive of the person’s personality type.",
      "research_paper": "assets/pdfs/35877f904befc83aa323dff783b495dd8231787f.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://storage.googleapis.com/kaggle-datasets-images/1831626/2988740/27fb5342b4c3cd9e782015d7b791bb87/dataset-cover.png?t=2021-12-30-19-31-49",
      "headline": "This project aims to lessen the reliance of self-report for personality typology through an artificial intelligence algorithm that can type people as one of the 16 MBTI types using an unedited writing sample by that person.",
      "project_id": "ded1e42ec9ef421c5cd42fdacb594176a55a3139",
      "expand": false,
      "tags": [
        "Pop Culture",
        "Psych",
        "Writing"
      ],
      "related_proj": [
        162,
        146,
        103
      ]
    },
    {
      "student_name": "Nicholas K.",
      "mentor_name": "Shreyas Muralidharan",
      "mentor_title": "PhD Candidate in Electrical and Electronics Engineering at Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/58cc40701fa3527a94a968779531266e739814b6.png",
      "domains": [
        "Environment"
      ],
      "project_title": "AI in Recycling",
      "project_desc": "The plastic pollution issue is an issue that has been plaguing the earth for the past half century and longer. Despite this, it has only been brought to light in the 21st century, and technology to help solve the issue is still new and largely undeveloped. Being someone who is passionate about environmental conservation, I wanted to create a tool that could be used to more easily sort recycling items. By using a machine learning algorithm, we have been able to create a tool that can detect what type of material an item is and determine whether it is recyclable or not. This tool can be applied to many different industry sectors, whether it is sorting items in a recycling plant, building it into an app for people to use day to day, or even aiding the plastic oceans cleanup effort. By creating this tool, we help do our part in protecting our planet and helping those who are less educated in the field to do so as well. We have approached the problem by using various machine learning algorithms to identify the material of items in a dataset. Using a KNN and CNN algorithm, we created a tool that became more and more accurate in identifying the material of an item. After testing the KNN and the CNN algorithm, we found that the CNN algorithm was much more accurate reaching a 68 % accuracy compared to 46% accuracy ",
      "research_paper": "assets/pdfs/fb2b02eb30f92383b188a126b2114e6d6d7fa527.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://img.vendingmarketwatch.com/files/base/cygnus/vmw/image/2022/03/16x9/SBX2022314_Recycling_Innovation_1536x1024.6231f4902dcf6.png?auto=format,compress&fit=fill&fill=blur&w=1200&h=630",
      "headline": "By using a machine learning algorithm, we have been able to create a tool that can detect what type of material an item is and determine whether it is recyclable or not.",
      "project_id": "d65aecd529c0729d9126735bb39fde11ccd3a9e4",
      "expand": false,
      "tags": [
        "Environmentalism",
        "Recycling",
        "Classification"
      ],
      "related_proj": [
        125,
        27,
        172
      ]
    },
    {
      "student_name": "Rohan S.",
      "mentor_name": "Christina Cheng",
      "mentor_title": "Materials Science and Engineering PhD Candidate at Stanford, ",
      "mentor_image": "assets/images/mentor_imgs/f13cf99371756efbf3731dd958f71a571ea1072d.png",
      "student_image": "assets/images/student_imgs/2429ae44ad0c64a6b0ea81ed9e9edeb9bc55b590.png",
      "domains": [
        "Education"
      ],
      "project_title": "Standardized Testing is it Really Effective?",
      "project_desc": "How effective is standardized testing in assessing students' knowledge? With how much standardized testing has been reinforced in our education, students should be properly assessed for their knowledge, and not be given tests that interfere with their academic learning. Under the federal, Every Student Succeeds act, schools must provide standardized tests. However, some of these standardized tests might not be doing what their intended purpose is and instead interfering with academic learning. had data from previous students on sat scores and GPAs from college and high school. I can use this data and plug it into different models to determine how well these sat scores determine GPA. I used different models to help determine and plot the results. To evaluate and determine how accurate these results are, I used three different error-calculating methods. When using linear regression to evaluate sat scores and first-year college GPA, there was a 0.532 mean percent error. This error seemed high for these two variables. There isn’t a strong enough correlation between sat scores and college GPA for it to be implemented in schools across the nation.",
      "research_paper": "assets/pdfs/99ea6674663522b4cb0bb9fbb4cfa1763ea8196d.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://marylandmatters.org/wp-content/uploads/2021/04/answer-sheet-1.jpeg",
      "headline": "How effective is standardized testing in assessing students' knowledge? With data from previous students on sat scores and GPAs from college and high school, I can use this data and plug it into different models to determine how well these sat scores determine GPA.",
      "project_id": "aac3c77f5f93d758213854796a0c04ba7bd27ae0",
      "expand": false,
      "tags": [
        "Education",
        "Standardized Testing"
      ],
      "related_proj": [
        80,
        77,
        159
      ]
    },
    {
      "student_name": "David S.",
      "mentor_name": "Roger Jin",
      "mentor_title": "Computer Science MS at MIT, ML researcher at Apple",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Journalism",
        "Language and Literature"
      ],
      "project_title": "Fake News Detection with BERT",
      "project_desc": "Fake news has become a prevalent issue in today’s\nsociety, with the ability to spread rapidly and\ncause harm through misinformation. In this paper,\nwe propose a fake news detection model using a\nfine-tuned BERT (Bidirectional Encoder Repre-\nsentations from Transformers) model. BERT is a\npowerful language model that has been successful\nin a variety of natural language processing tasks.\nWe fine-tune the BERT model on a dataset of real\nand fake news articles from Kaggle, and evaluate\nits performance on a holdout test set. Our results\nachieve an accuracy of 0.574 on the fake news\ndetection task.",
      "research_paper": "assets/pdfs/b62726ea9e02a33a78c7a323156a6fc90ba316e4.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://about.fb.com/wp-content/uploads/2020/06/117592988_613376329317475_1014520698677157145_n.jpg?resize=890%2C501",
      "headline": "In this paper, we propose a fake news detection model using a fine-tuned BERT (Bidirectional Encoder Representations from Transformers) model.",
      "project_id": "6f3fc38d7da0d7ab52b86f567f619edd9d3239a7",
      "expand": false,
      "tags": [
        "News",
        "Fake News",
        "BERT"
      ],
      "related_proj": [
        81,
        84,
        57
      ]
    },
    {
      "student_name": "Ayrton S.",
      "mentor_name": "Bryce Johnson",
      "mentor_title": "Computer Science Stanford Alum, Industry Data Analytics and Business Consultant at Oliver Wyman",
      "mentor_image": "assets/images/mentor_imgs/a0c3954a9a3a7373b7ac9acb5e36983a19efc18c.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Predicting Repeat Purchases in E-Commerce",
      "project_desc": "As companies move their business online, the e-commerce space continues to develop as a rapidly growing industry. Marketing and upselling are crucial aspects of profitability. Therefore the objective of this investigation is to determine how the price, along with other features, of items sold on Amazon affect the likelihood of a customer making a repeat purchase. The underlying problem here is identifying more efficient upselling strategies for retailer companies. These findings may potentially help a retail company determine the likelihood of a customer purchasing another item based on their first purchase, and then decide how to upsell based on that information. Customer behavior and motivation is a crucial aspect of repeat purchases and original purchases in general, this notion has to be understood before attempting to answer the question. Identifying which features held the most weight in predicting repeat purchases was also a crucial aspect in determining if customers will buy again. In general it could be deduced that price, product brand, and product rating carry a great deal of weight when it comes to customer’s making repeat purchases.",
      "research_paper": "assets/pdfs/4de402cc2048bee6f7bd4a7b943d10f80d5e8d00.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://store.hp.com/app/assets/images/uploads/prod/ecommerce-vs-online-marketplace1603983748724190.jpg",
      "headline": "The underlying problem here is identifying more efficient upselling strategies for retailer companies. These findings may potentially help a retail company determine the likelihood of a customer purchasing another item based on their first purchase, and then decide how to upsell based on that information.",
      "project_id": "acf2bdaf2675d101299af10dc1b6e5d411d0f085",
      "expand": false,
      "tags": [
        "Economics",
        "Consumer Behavior"
      ],
      "related_proj": [
        122,
        22,
        2
      ]
    },
    {
      "student_name": "Bobby B.",
      "mentor_name": "Tomer Arnon",
      "mentor_title": "MS Engineering at Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/7e0bc07efda21080cec4ac7af76393602f4f7ddb.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Predicting the Price of New York City Airbnbs",
      "project_desc": "How can one predict the price of a New York City Airbnb? We are trying to create a machine learning model that can predict the price of a NYC Airbnb given some factors with high accuracy. There are many factors that contribute to pricing an Airbnb. This model is important because it can help people list their Airbnb for a fair price and help a renter determine if a listing price is fair. We first found a dataset containing the price of New York City Airbnbs and many features about these houses. This dataset also contained many factors about the houses that could be important for predicting prices. We then condensed the data into only the data that we thought were important. We compared several regression models to see which would predict the price best. The random forest regressor predicted the price with about 27% away from the price on average. The SGD regression predicted the price with an error of about 42%. Our model wasn’t very accurate when predicting the price. To accurately predict the price, more factors about the housing are most likely needed. Some of these could be square footage, bathrooms, and bedrooms. Though, these weren’t on the dataset we used.",
      "research_paper": "assets/pdfs/86e08cacbf82fb0717d19671eae734ee4d7bffbc.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://vrmintel.com/wp-content/uploads/2022/03/shutterstock_1733117681-1000x600.jpg",
      "headline": "How can one predict the price of a New York City Airbnb? We are trying to create a machine learning model that can predict the price of a NYC Airbnb given some factors with high accuracy. ",
      "project_id": "9c6db7729f0bd3434f86a8496bbea560d83db38b",
      "expand": false,
      "tags": [
        "Economics",
        "New York",
        "Prediction"
      ],
      "related_proj": [
        2,
        22,
        121
      ]
    },
    {
      "student_name": "Aishwaroopa N.",
      "mentor_name": "Tomer Arnon",
      "mentor_title": "MS Engineering at Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/8721cb300b508dd51509c6f7fe23c34c1c8bceff.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Using EEG to Detect Eye Movement",
      "project_desc": "In this paper, we show that it is possible to use EEG data to detect eye movement using machine learning. By recognizing eye movement through EEG results, our goal is to help individuals with disabilities better control object movement and perform daily activities independently. This is especially important as many disabled individuals rely on assistance from others for their daily needs, which can be burdensome for the person providing help. To achieve these objectives, we trained different machine learning models using a data set of eye-state classification from Kaggle. We analyzed the results to assess the accuracy of a KNN (K Nearest Neighbors) model. With the model achieving an accuracy of 95.23% in detecting eye movement in patients. These findings suggest that the model could be effectively utilized in the future, with further research to assist individuals with disabilities. Overall, our research suggests that it is possible to recognize eye movement through EEG results reliably. Further research in this area could lead to the development of more effective and personalized interventions for individuals with poor hand-eye coordination.",
      "research_paper": "assets/pdfs/d2d365ee73270aa118e835366612c9d3087861e3.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.rochester.edu/newscenter/wp-content/uploads/2023/02/fea-fixational-eye-movements.jpg",
      "headline": "In this paper, we show that it is possible to use EEG data to detect eye movement using machine learning. By recognizing eye movement through EEG results, our goal is to help individuals with disabilities better control object movement and perform daily activities independently.",
      "project_id": "93f69f6e27f7f28be4488f702851296b8fccb230",
      "expand": false,
      "tags": [
        "Physiology",
        "Eyes"
      ],
      "related_proj": [
        134,
        26,
        178
      ]
    },
    {
      "student_name": "Roneet D.",
      "mentor_name": "Tony Rodriguez",
      "mentor_title": "PhD Candidate in Astrophysics at Caltech",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Findings Provided by a Machine Learning Clustering Model of Stellar Kinematics in Star Clusters Can be Used to Identify Runaway Stars",
      "project_desc": "Star clusters dissipate as internal and external gravitational interactions weaken its tidal radius, causing smaller groups of stars to split apart in several directions while pushing other stars along with it. The need for developing a machine learning model that can detect these runaway stars is established in this research as it attempts to expand the domain of astrophysics to help astrophysicists understand tidal tail formation and the theoretical dark matter subhalos’ effects on celestial bodies. Data from the ESA’s Gaia DR 3 was utilized and noise was removed through data cleaning with 4 parameters: RAdeg, DEdeg, PM, and Distance that were then factored into a K-means clustering algorithm, which created the star clusters. These 4 factors were subsequently compared and analyzed in various scatterplots. The findings revealed the model’s success in identifying potential candidates of runaway stars by confirming the existence of high proper-motion stars amongst other more similar proper-motion stars and the ability for stars to be separate by vast distances across their cluster, indicating possible tidal tails. Future improvements to work around various limitations encountered throughout this research focus on the development of a hybrid clustering-nueral network model to automatically detect and predict the likelihood of runaway stars from their clusters. ",
      "research_paper": "assets/pdfs/68cece4d99993ebd66be937d4dcfd7348ff521a5.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://cdn.mos.cms.futurecdn.net/DyjQV2xGR43rrbfVcn8wqU-320-80.jpg",
      "headline": "The need for developing a machine learning model that can detect these runaway stars is established in this research as it attempts to expand the domain of astrophysics to help astrophysicists understand tidal tail formation and the theoretical dark matter subhalos’ effects on celestial bodies.",
      "project_id": "f78d53783940a77db8cf626d3031cc95f5ad469b",
      "expand": true,
      "tags": [
        "Astronomy",
        "Stellar Classification",
        "Kinematics"
      ],
      "related_proj": [
        151,
        168,
        3
      ]
    },
    {
      "student_name": "Joseph N.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Arts and Music"
      ],
      "project_title": "Sketch Recognition using Artificial Intelligence",
      "project_desc": "This paper is about an AI project that helps people learn about animals. The AI can recognize hand-drawn sketches of animals and other objects and provide information about them. I was motivated to start this project because of my interest in art and wanted to build an AI system that could learn to detect art created by humans. I used different supervised classification algorithms such as Logistic Regression, Random Forest and Ridge Classifier to train our model. Random Forest classifier performed the best among all the models we used with an accuracy of 72.8%. I used Python and used Google’s Quickdraw to collect images and data.",
      "research_paper": "assets/pdfs/86415b2f98bfcd8cf6ce65bdc2a7c5e0e6fc82f4.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "github": "https://drive.google.com/drive/folders/1lBCXrXiEam_gs7-SeL25IwB6Xa6_8qc5",
      "graphic_link": "https://cdn.psychologytoday.com/sites/default/files/styles/article-inline-half-caption/public/field_blog_entry_images/2018-09/sketch_2.jpg?itok=0ZpcuMDD",
      "headline": "This paper is about an AI project that helps people learn about animals.",
      "project_id": "1a5ced28013c5a496d372a1278286642217a817d",
      "expand": false,
      "tags": [
        "Visual Arts",
        "Google",
        "Classification"
      ],
      "related_proj": [
        144,
        62,
        176
      ]
    },
    {
      "student_name": "Rohan T.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Brain Cancer Detection",
      "project_desc": "Through the rapid destruction of brain cells, malignant brain tumors can be one of the most dangerous forms of cancer that exist. Current methods for determining the presence and type of brain tumor in a given patient’s MRI scan can oftentimes be inefficient and are prone for error. By using a machine learning algorithm, the error in these classifications is reduced significantly, and the process is made much more efficient. To approach this problem, we have implemented an Inception Resnet, which is a pretrained Convolutional Neural Network. The final model is able to determine the presence of a possible brain tumor as well as distinguish the type between three different classes. These classes include pituitary, meningioma, and glioma. Pituitary tumors are the most harmless of the three as they are noncancerous. The five year survival rate for a patient with a pituitary tumor is 97%. This contrasts greatly with both glioma and high grade meningioma tumors which are life-threatening [2]. It should be noted, however, that high grade meningioma tumors are very rare. Along with these three classes, a separate control class with no brain tumors was also used. After training the model on these four classes, the results were found to be impressive, as the final accuracy was at 96.7%. ",
      "research_paper": "assets/pdfs/1af78eeff767deaeb67b548a0580dff075490f4e.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://medicine.wustl.edu/wp-content/uploads/GlioblastomaBranScans.jpg",
      "headline": "Current methods for determining the presence and type of brain tumor in a given patient’s MRI scan can oftentimes be inefficient and are prone for error. By using a machine learning algorithm, the error in these classifications is reduced significantly, and the process is made much more efficient.",
      "project_id": "27117f99bd699eb87adf3d43adad7cbccafacd69",
      "expand": false,
      "tags": [
        "Brain Cancer",
        "CNNs",
        "Cancer"
      ],
      "related_proj": [
        185,
        98,
        156
      ]
    },
    {
      "student_name": "Yuna S.",
      "mentor_name": "Shreya Parchure",
      "mentor_title": "MD-PhD student at UPenn",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Education"
      ],
      "project_title": "A new style of teaching: Exploring the benefits of visual language learning",
      "project_desc": "This project was designed to investigate the efficiency and likelihood of people learning a new language through means of an AI that would help translate objects in day to day life to a language that the user would want to learn. This specific project would help to benefit a better understanding of languages and enable the user to have better connection and communication skills with others. The overall approach taken to achieve this goal came in the form of two parts: object detection and natural language processing (NLP). The first part contained training the AI on certain categories of objects which would then be put through the second part of the AI, the translation. The user would also be able to interact with the AI by inputting their own data and selecting a certain language they would like to learn. The AI was seen to have around a 35% accuracy rate in determining the correct image but had a few errors when encountering the translation side of things. Regardless, some important conclusions were that the AI was successful in being able to translate languages and would certainly help people in their language learning endeavors. ",
      "research_paper": "assets/pdfs/caba5a6db17f372e1f4acf82fe5daa4978c509e0.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://static1.makeuseofimages.com/wordpress/wp-content/uploads/2022/11/Google-Translate-on-an-Android-phone.jpg",
      "headline": "This project was designed to investigate the efficiency and likelihood of people learning a new language through means of an AI that would help translate objects in day to day life to a language that the user would want to learn.",
      "project_id": "45f834c7383730d93c58656b5897988425ec17da",
      "expand": false,
      "tags": [
        "Modern Languages",
        "Natural Language Processing",
        "Education"
      ],
      "related_proj": [
        80,
        86,
        77
      ]
    },
    {
      "student_name": "Vadim Y.",
      "mentor_name": "Darnell Granberry",
      "mentor_title": "MEng Computer Science from MIT, current Machine Learning Engineer",
      "mentor_image": "assets/images/mentor_imgs/d2773fffd8d24451e14181829c0c3ba1adc8bbf6.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Predicting Precipitation and Other Weather Conditions With Logistic Regression and Random Forest Classifiers",
      "project_desc": "Nowadays, we predict the weather to get a better understanding of the days to come. An issue we have with our current algorithms is the inability to predict weather accurately for any time range further than ten days. Our intentions with this project were to create a model that could get an accuracy within 5% of our current weather prediction algorithms. We wanted the model to be accurate in both short-term and long-term conditions. Our current algorithms, while highly accurate in the short term, struggle in the long-term timespan. The importance of accuracy in this model can be boiled down to two main points; convenience, for planning activities, and safety, as perhaps an early warning system in the event of a possibly harmful or fatal weather event. Our method of achieving these goals was to  first use a logistic regression model, and later a random forest classifier. In the end we reached a highest accuracy of 83%, which fell just short of our goal of 85%. Despite the inaccuracy, our model showed promising results, and if we had more time or a larger dataset, the accuracy could have been higher. In the end, the results seemed more to be a proof of concept, rather than a demonstration of accuracy and capability. ",
      "research_paper": "assets/pdfs/fc0090c1f939855ee18545e60d1f986db0fb6948.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://s.w-x.co/West_Plains_1107a.jpeg?crop=16:9&width=320&format=pjpg&auto=webp&quality=60",
      "headline": "An issue we have with our current algorithms is the inability to predict weather accurately for any time range further than ten days. Our intentions with this project were to create a model that could get an accuracy within 5% of our current weather prediction algorithms.",
      "project_id": "32d254828e54221df930fa94f6c3b043b0ba2f34",
      "expand": true,
      "tags": [
        "Weather",
        "Earth Systems"
      ],
      "related_proj": [
        54,
        33,
        145
      ]
    },
    {
      "student_name": "Michael T.",
      "mentor_name": "Sean Konz",
      "mentor_title": "Computer Science MS from Stanford, Software Engineer",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "A Methodology for Learning Airplane Descent Patterns",
      "project_desc": "In this study, we trained a plane to land without human assistance. We did this using reinforcement learning, a type of machine learning where the agent (the AI “pilot” in this case) runs iterations of the task and adjusts its methodology depending on this interaction with its environment. This agent-environment interaction is called the Markov Decision Process, and is the heart of reinforcement learning. This is an important problem to solve as reinforcement learning applications like these can be used to improve the safety and quality of life for people who are regularly traveling. Additionally, the fact that reinforcement learning can be used to solve problems like these implies further applications in complex fields like flight and other modes of transportation. To figure out how to train a plane to land, we used OpenAI Gym to build a simulated environment where we tested different landing maneuvers. Initially, we experimented with optimizing the angle of attack for variables like time taken to land and descent rate, but ultimately found it most effective to optimize for having the lowest velocity when the plane reaches the ground. Further research could potentially experiment with how different variables such as speed, acceleration, and angles of the left and right ailerons affect the plane’s landing.",
      "research_paper": "assets/pdfs/171d4733d4c0de99d3447fde902d34964ec28f08.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://assets-global.website-files.com/5b27a494dc3d312fe5c01da8/6302b517d3a4b32950df5104_Screen%20Shot%202022-08-21%20at%205.43.15%20PM.png",
      "headline": "In this study, we trained a plane to land without human assistance.",
      "project_id": "41eff1ff1ea4f09dbcc0b372f9b3661ab1771abb",
      "expand": false,
      "tags": [
        "Engineering",
        "Autonomous Vehicles",
        "Reinforcement Learning",
        "Aerospace Engineering"
      ],
      "related_proj": [
        175,
        180,
        166
      ]
    },
    {
      "student_name": "Vinati P.",
      "mentor_name": "Samuel Kwong",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Stroke Susceptibility Prediction Using Artificial Neural Networks",
      "project_desc": "Strokes are among the top five leading causes of death in the United States, with over 140,000 patients dying each year in America alone. Our research question is centered around predicting stroke susceptibility and determining which demographic and medical factors significantly increase or decrease that risk. Stroke research in the past has been centered around determining the presence of a stroke using CT imaging after it has occurred; our goal was to determine likelihood of stroke prior to a stroke occurrence in order to prepare patients for and proactively address stroke risk. As such, a deep learning model was developed for stroke prediction and tested on a dataset of 5110 samples. Hyperparameters such as operating point, learning rate, batch size, and epoch value were tuned to maximize model efficiency. The model had an overall accuracy of 0.74, accurately determining 660/954 stroke negative samples and 41/54 stroke positive samples, producing recalls of 0.68 and 0.76 respectively. Overall, it was found that age played the largest role in stroke susceptibility with a correlation value of 0.25. While it was found that medical factors played a larger role in stroke proneness than lifestyle factors, the correlation between average glucose level, heart disease, hypertension, and stroke risk and the correlation between marriage status and stroke risk remained similar at 0.13 and 0.11. Our results indicate the potentiality of the use of AI in determining stroke risk based solely on collected medical and lifestyle data.",
      "research_paper": "assets/pdfs/913063f02978378e1041e8ca4c486bc05644d2dc.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://vitals.sutterhealth.org/files/2020/05/iStock-1168179082.jpg",
      "headline": "Our research question is centered around predicting stroke susceptibility and determining which demographic and medical factors significantly increase or decrease that risk. Our results indicate the potentiality of the use of AI in determining stroke risk based solely on collected medical and lifestyle data.",
      "project_id": "2aa71ce9beb4d3442eb8fd6b29b15511db67485f",
      "expand": false,
      "tags": [
        "Strokes",
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "related_proj": [
        50,
        42,
        61
      ]
    },
    {
      "student_name": "Samya C.",
      "mentor_name": "Shreya Parchure",
      "mentor_title": "MD-PhD student at UPenn",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Comparing Machine Learning Models to Determine Which is Most Effective at Detecting Brain Tumors",
      "project_desc": "In this project, by using machine learning techniques to analyze various brain tumor scans, the goal was to determine which techniques are the most efficient and accurate in determining the presence of brain tumors. Specifically, the goal was to adequately process a dataset to determine if brain tumors can be detected with more than 75% accuracy using machine learning. The chosen dataset includes just over 250 axial MRI brain scans, thus providing a sufficient dataset to properly analyze. This research is most applicable to the healthcare field, specifically relating to the work done by neurologists and radiologists. If the most efficient and accurate way to detect brain tumors is not determined soon, individuals may have to continue waiting longer than necessary for their results. In this project, K-Nearest Neighbors, Decision Trees, and Muti-Layer Perceptron Models were compared to determine which machine learning algorithm is most effective at determining the presence of a brain tumor in an axial brain scan.",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.radiology.ca/wp-content/uploads/2022/09/mri-brain_web.jpg",
      "headline": "By using machine learning techniques to analyze various brain tumor scans, the goal was to determine which techniques are the most efficient and accurate in determining the presence of brain tumors.",
      "project_id": "a261ae580ac91826850be5ebf96dec1675450b5d",
      "expand": true,
      "tags": [
        "Cancer",
        "Brain Cancer",
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "related_proj": [
        13,
        6,
        8
      ]
    },
    {
      "student_name": "Akshay N.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/2884d597b1f0f73604aad904746eb3b128787b3b.png",
      "domains": [
        "Physics",
        "Engineering"
      ],
      "project_title": "Vibration Analysis",
      "project_desc": "This paper covers differences in accuracies of an artificial intelligence model that classifies different sets of sensor outputs (in this case, machinery shaft vibrations) collected by sensors into varying levels of weight on the shaft. The model uses a neural network to categorize the data, and there are varying accuracies depending on weights at the end of the shaft. This paper goes over the causes for these changes and patterns one can notice by going through the results.",
      "research_paper": "assets/pdfs/f57a1c896a116bbec9d4fc9f38e4c99dd8d54c74.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://www.datocms-assets.com/38028/1667940979-industrial-iot-sensor.png",
      "headline": "This paper covers differences in accuracies of an artificial intelligence model that classifies different sets of sensor outputs (in this case, machinery shaft vibrations) collected by sensors into varying levels of weight on the shaft.",
      "project_id": "1db4612635034101fc7503af5cac8451fcc6ffb5",
      "expand": false,
      "tags": [
        "Engineering",
        "Hardware",
        "Physics"
      ],
      "related_proj": [
        96,
        181,
        166
      ]
    },
    {
      "student_name": "Tarun S.",
      "mentor_name": "Katie O'Nell",
      "mentor_title": "PhD Student at Dartmouth, Brain and Cognitive Sciences BS from MIT, Ethics+Social Sciences project and course developer at Inspirit AI",
      "mentor_image": "assets/images/mentor_imgs/3df727e01bd6076d80bdefdd1c47288c998e0f7e.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "PoliSci and Law",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Exposing Undercounts in the Census through Regression Modelling",
      "project_desc": "Undercounts in the Census are notoriously difficult to measure, but are necessary to understanding the extent of systemic divides within America. Although many community leaders have proposed that language barriers pose significant obstacles to Census outreach, this paper explores the viability of using predictive models to quantify the extent the role language plays. By using a multivariable regression model trained on student data from the Civil Rights Data Collection, we concluded that Limited English Proficient (LEP) Asian populations were undercounted by over 98,000 people, LEP Native Hawaiian/Pacific Islander populations were undercounted by over 73,000 people, and LEP White populations were undercounted by over 164,000 people.  However, biases in the dataset made the results for other ethnic groups unreliable, indicating that regression modeling should be used as a tool for identifying areas of improvement in the Census rather than producing an exact estimate of the population. Our data suggests a strong correlation between language proficiency and Census undercounts, and the Census Bureau could ideally use the results to create targeted language outreach programs for specific states.\n\t\n",
      "research_paper": "assets/pdfs/c7be5f7cc58b20506e41a7033a900991ab642610.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "github": "https://github.com/tshah3500/Exposing-Undercounts-in-the-Census-through-Regression-Modeling",
      "graphic_link": "https://npr.brightspotcdn.com/legacy/sites/kwmu/files/202002/021320_NT_census_2020.jpg",
      "headline": "Although many community leaders have proposed that language barriers pose significant obstacles to Census outreach, this paper explores the viability of using predictive models to quantify the extent the role language plays.",
      "project_id": "41b97acb4bd9cc99bf474f62c6ae41e5388af259",
      "expand": false,
      "tags": [
        "Government Systems",
        "Inequality",
        "Racial Equality",
        "Modern Languages"
      ],
      "related_proj": [
        47,
        35,
        165
      ]
    },
    {
      "student_name": "Srinivas S.",
      "mentor_name": "Nima Leclerc",
      "mentor_title": "UPenn PhD in Electrical Engineering",
      "mentor_image": "assets/images/mentor_imgs/f301042e741b8df589841eaf068176c844806187.png",
      "student_image": "assets/images/student_imgs/6b5487f76029edef5909aa6bbbc58a923af9057c.png",
      "domains": [
        "Healthcare and Biology",
        "Psychology"
      ],
      "project_title": "AI Detection of Emotions through EEG",
      "project_desc": "Electroencephalogram (EEG) brain waves are an extremely important resource for clinical analysis and diagnosis of mental health and other psychiatric diseases. They help current diagnosis of these diseases, as clinics use the external behavior of a patient and the psychiatrist’s qualitative reasoning to determine what disease the patient has. However, this approach does not examine internal activity in the brain, and is not uniformly quantifiable. EEG provides a means for psychiatrists to do this, however, this process can be complicated because the variability in EEG signals. Machine Learning (ML) provides a means to analyze these complex data. A computational tool that can recognize patterns in EEG signals would be a valuable starting point in diagnosis psychiatric disorders. We proceeded to develop a recurrent neural network (RNN) model structure that could analyze an individual’s brain waves and predict their emotional condition. We used the EEG Brainwave Dataset: Feeling Emotions dataset, which contains various statistical features of the EEG brain waves of two different test subjects as they were experiencing different emotions. We extracted frequency domain features, converted them to the time domain, and filtered them. We trained a gated recurrent unit (GRU) model from these filtered and attained a 96.95% validation accuracy.",
      "research_paper": "assets/pdfs/94cf93c69e828accbf886dbbc5b16d23acb84c68.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/srinisriram/AI_Detection_EEG",
      "graphic_link": "https://images.hindustantimes.com/img/2022/08/09/550x309/pexels-andrea-piacquadio-3812743_1660054870963_1660054906342_1660054906342.jpg",
      "headline": "We proceeded to develop a recurrent neural network (RNN) model structure that could analyze an individual’s brain waves and predict their emotional condition.",
      "project_id": "dfcd5decb9debec976f2be8a62ace6493354126c",
      "expand": false,
      "tags": [
        "Facial Emotion",
        "Mental Health",
        "Psych"
      ],
      "related_proj": [
        183,
        14,
        90
      ]
    },
    {
      "student_name": "Zhou C.",
      "mentor_name": "Ross Greer",
      "mentor_title": "PhD Candidate in Electrical and Computer Engineering, Research at MIT and Berkeley, Engineer at Tesco",
      "mentor_image": "assets/images/mentor_imgs/c995a8ff1784ed78d976edf101087772c99899fa.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Arts and Music"
      ],
      "project_title": "T-RECSYS+: An Improved Music Recommendation System",
      "project_desc": "A recommendation system is a type of filtering system that predicts a user's preferences for a specific item. Its purpose is to suggest items that the user might find appealing. In our research, we build a music recommendation system to make prediction of users' listening preference. Our system extends the previous T-RECSYS algorithm which uses a hybrid of content-based and collaborative filtering as input to a deep learning classification model. We enhance the performance of this algorithm by incorporating the latest Spotify API, which provides access to 11 music features including danceability, liveness, tempo and so forth. Additionally, we leverage more advanced deep learning models to achieve a higher level of precision and accuracy in our recommendations. In detail, we promote the precision scores from the original 88% to the current over 95%.",
      "research_paper": "assets/pdfs/97fabc882c91c7aaefda7301ad254f0ea223d3af.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/zcj0125/T-RECSYS-An-Improved-Music-Recommendation-System",
      "graphic_link": "https://images.unsplash.com/photo-1616356607338-fd87169ecf1a?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxzZWFyY2h8Mnx8bXVzaWMlMjBzdHJlYW1pbmd8ZW58MHx8MHx8fDA%3D&w=1000&q=80",
      "headline": "In our research, we build a music recommendation system to make prediction of users' listening preference. ",
      "project_id": "2f2df5aaa8a8cb7605c3ba9894d7babb208cdaff",
      "expand": false,
      "tags": [
        "Music",
        "Deep Learning",
        "Pop Culture"
      ],
      "related_proj": [
        62,
        144,
        76
      ]
    },
    {
      "student_name": "Claire L.",
      "mentor_name": "Katie O'Nell",
      "mentor_title": "PhD Student at Dartmouth, Brain and Cognitive Sciences BS from MIT, Ethics+Social Sciences project and course developer at Inspirit AI",
      "mentor_image": "assets/images/mentor_imgs/3df727e01bd6076d80bdefdd1c47288c998e0f7e.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Psychology"
      ],
      "project_title": "What Data is Needed to Accurately Determine Someone's Mental Health?",
      "project_desc": "As the demand for mental healthcare increases, more people have turned to mental health chatbots and other AI-powered teletherapy options. While these chatbot applications expand access to care that is often expensive and stigmatized, they also pose some cybersecurity risks. Some chatbot applications use phone sensors and other device data to make predictions about the severity of a patient’s mental illness. While this makes these predictions more accurate, they might compromise user trust. In this paper, we are trying to find which data we should collect from the user’s phone to minimize the amount of data being collected but also collecting adequate information to predict their mental health score. This is important because users are not fully trusting the mobile apps since they are afraid the apps might track all of their personal information. If we determine which data is actually necessary, we can build better user trust while maintaining the efficacy of a chatbot. The overall approach was to first collect the data that was deemed useful, such as education, sensing, survey, and Ecological Momentary Assessment (EMA) data, and then found how the accuracy of the scores would decrease when each category of data collected was removed to determine the most important category to collect, as well as what combination of categories yielded the highest accuracy. The most significant result was that when we didn’t include the data from the surveys, the accuracy went down significantly. The major conclusions are that not every detail from a user’s phone will need to be collected in order to yield accurate results, and in fact, simply asking users about their daily experiences allows for far more accurate results than on-device measures.",
      "research_paper": "assets/pdfs/d8fc18b02f8d8586076f871696b52235878753ea.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.verywellmind.com/thmb/2SiXj67NAsvcBc_odmAEYimD0CM=/2120x0/filters:no_upscale():max_bytes(150000):strip_icc()/GettyImages-14213709881-a59ff3a3cc8d44149117df9a658f46e4.jpg",
      "headline": "If we determine which data is actually necessary, we can build better user trust while maintaining the efficacy of a chatbot.",
      "project_id": "2e324cf18859058cfb75c0543cf269a2a3ca9042",
      "expand": false,
      "tags": [
        "Mental Health",
        "Data and Privacy"
      ],
      "related_proj": [
        59,
        21,
        137
      ]
    },
    {
      "student_name": "Khushi G.",
      "mentor_name": "Joseph Vincent",
      "mentor_title": "Aerospace Engineering PhD Candidate at Stanford",
      "mentor_image": "assets/images/mentor_imgs/808f2a4f0dad138963d86e72fa8a6bca1820e194.png",
      "student_image": "assets/images/student_imgs/6e0a246bb918803da3a91ee66eb2fcecee65b1a4.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Dental Segmentation of The Mandible Using AI",
      "project_desc": "Dental segmentation can help with decision-support issues in medical diagnosis, such as human identification and maxillofacial surgery, as well as orthodontic therapy, implant planning, and other issues. This leads to my research question; “How do I automate the task required to diagnose a dental problem during a dental visit?” The approach to solve this was to train an AI with images pre-segmented by experts and then test it. This process is also called image segmentation. The AI had eventually gotten a test accuracy score of  0.940 and test IoU score of  0.904. The closer to one that these scores are the better. Pixel accuracy and Iou score were surprisingly similar. I think that this is a good thing because it means both seem pretty accurate. The AI has its flaws but it can segment images when given images",
      "research_paper": "assets/pdfs/00015d8f77a3da9efae363863da292ec833464a5.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://www.absolutedental.com/wp-content/uploads/2012/10/patient-at-the-dentists-office.jpg",
      "headline": "Dental segmentation can help with decision-support issues in medical diagnosis, such as human identification and maxillofacial surgery, as well as orthodontic therapy, implant planning, and other issues. This leads to my research question; “How do I automate the task required to diagnose a dental problem during a dental visit?”",
      "project_id": "d15ec1b5bee0b0f29862e6546548a232928f0694",
      "expand": false,
      "tags": [
        "Dental Health",
        "Computer Vision"
      ],
      "related_proj": [
        117,
        119,
        115
      ]
    },
    {
      "student_name": "Nicholas T.",
      "mentor_name": "Sophia Barton",
      "mentor_title": "Computer Science MS from Stanford",
      "mentor_image": "assets/images/mentor_imgs/56c2b2fb8dc613f5a0a7ae222a5ee595734e2950.png",
      "student_image": "assets/images/student_imgs/5253bdab3dc2db0484dd21f03f089541ee6bd832.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Cardiac Auscultation: Metrics of Smartphones and Digital Stethoscopes ",
      "project_desc": "The objective of our research is to figure out how feasible and accurate a mobile device solution to cardiac ascultation is, compared to a digital stethoscope. Having to pay to go to a doctor's office and pay for a medical professional to use a stethoscope is costly and inconvenient. Having a mobile solution that is cheap and uses a medium that is widespread will make diagnoses more accessible. We used a convolutional neural network-based solution, which used the heart sound audio, collected with a digital stethoscope and smartphone, graphed out on a spectrogram for input. The model trained on the smartphone data typically performed 15% worse than the model trained on stethoscope data in terms of accuracy. The hardware technology in phones is still not advanced enough to reliably diagnose with machine learning based off of these results alone.\n",
      "research_paper": "assets/pdfs/272137c63a2a27199fd01299e655c28de3a1e8ac.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://riester.de/media/pages/en/products/stethoscopes/advanced-digital-auscultation/modules/image/517a39dfbd-1674381317/riester-dr-patient-ri-sonic-289-web-1440x-q80.jpg",
      "headline": "The objective of our research is to figure out how feasible and accurate a mobile device solution to cardiac ascultation is, compared to a digital stethoscope.",
      "project_id": "00976cbc8acc933fdb58db38dd5af801a0a7cfc9",
      "expand": false,
      "tags": [
        "Heart Disease",
        "Social Justice"
      ],
      "related_proj": [
        184,
        143,
        117
      ]
    },
    {
      "student_name": "Rayyaan O.",
      "mentor_name": "Linda Banh",
      "mentor_title": "Electrical Engineering MS from Stanford with professional AI Certificate, Engineer at Microsoft",
      "mentor_image": "assets/images/mentor_imgs/3c9d0f6a41bcd9e11d20529cfd74449f29a85bb9.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Facial Intoxication Detector with AI",
      "project_desc": "Drunk driving is a very widespread problem, causing many casualties and millions of dollars in insurance and damages per year. In 2020, despite the COVID19 pandemic greatly reducing road traffic, about 32 people died per day in the US alone from DUI (Driving under Influence)(nhtsa.gov). Existing solutions such as police testing drivers suspected to be under influence is simply impractical considering the number of possible intoxicated drivers. With more and more people gaining access to a vehicle, it is crucial that more effective strategies be developed to detect and combat drunk driving. The purpose of this study is to analyze whether Artificial Intelligence can be used to more effectively detect and prevent drunk driving and if a Machine Learning models like Logistic Regression and Decision Tree can be more accurate than a human police officer. To address this question, a dataset named drunkImagesWebp was used. Two machine learning algorithms, Logistic regression and decision tree were then trained on this dataset with facial image data of intoxicated people to accurately predict the sobriety of humans based on facial cues. After testing this model, it became clear that both Logistic regression and decision tree models can indeed accurately test a driver for signs of intoxication with well over 90% accuracy compared to human-administered tests with can only hit up to around 75% accuracy. By comparison, both the Logistic regression and Decision tree algorithms detected intoxication with 96% accuracy. This paper shows the potential AI can have in creating an automated solution to detecting and ultimately preventing drunk driving. ",
      "research_paper": "assets/pdfs/66093cba3cd8fdecc4a2c144e9d017ff6e3c1fe5.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://i.kinja-img.com/gawker-media/image/upload/c_fit,f_auto,g_center,q_60,w_645/8b1709d78c0ce9d3251d332796033ce7.jpg",
      "headline": "The purpose of this study is to analyze whether Artificial Intelligence can be used to more effectively detect and prevent drunk driving and if a Machine Learning models like Logistic Regression and Decision Tree can be more accurate than a human police officer. ",
      "project_id": "bbb3616167b2e8749aefac1ccdc0ab267936c018",
      "expand": false,
      "tags": [
        "Social Justice",
        "Drugs",
        "Drug Overdose",
        "Logistic Regression"
      ],
      "related_proj": [
        56,
        9,
        143
      ]
    },
    {
      "student_name": "Arnav G.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/0e566202b048f3cbbcf8bbb932afbd5f1dd6de25.png",
      "domains": [
        "Healthcare and Biology",
        "Environment"
      ],
      "project_title": "Modeling The Impact of Electric Vehicle Adoption on NO2 Levels Using Machine Learning: A Predictive Analysis",
      "project_desc": "Air pollution, notably nitrogen dioxide (NO2), poses severe health and environmental risks. The main sources of NO2 are related to transportation and stationary fuel combustion sources [U.S Environmental Protection Agency (EPA), 2018]. With a global shift to electric vehicles (EVs), their potential to mitigate pollution is becoming prominent. In India, where air quality concerns are paramount, understanding the tangible benefits of EV adoption is crucial. The research question explores whether increasing EVs adoption can visibly reduce NO2 levels. The approach combines use of various supervised machine learning models and simulations to understand the impact of socio-economic factors,  adoption of electric vehicles, energy generations and emission factors on the NO2 levels and to  create a model to predict the level of NO2.  The created model will be used to estimate the impact different EV adoption levels can have on NO2. To train the model, data from 49 US states along with 5 Indian states where EV sales are comparable and the NO2 data was available has been used. Among diverse algorithms tested, the Neural Network turns out to be the best which predicts NO2 levels with a R2 value of 0.82. Simulations, varying EV numbers while holding other factors steady, demonstrated a clear link between higher EV count and lower NO2 levels. The results not only validate the potential of electric vehicles as catalysts for improving air quality but also highlight their crucial role within the Indian context, especially in the state of  Delhi which we have used as our case study. The empirical evidence underscores the pivotal contribution that EVs could make to enhancing Delhi’s air quality profile. In conclusion, the research provides substantial evidence that the widespread adoption of electric vehicles holds promise as an effective strategy to reduce NO2 levels, thereby laying the foundation for a more sustainable and healthier urban environment.",
      "research_paper": "assets/pdfs/4f9bc050f99c17580d1629c96369f7207f70310c.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://cloudfront-us-east-2.images.arcpublishing.com/reuters/W3PJTWEE7FJGFKEC72PMPFIWD4.jpg",
      "headline": "Air pollution, notably nitrogen dioxide (NO2), poses severe health and environmental risks. The research question explores whether increasing EVs adoption can visibly reduce NO2 levels.",
      "project_id": "40215bb18dab3df076b529d0cb7079e4782d2aa9",
      "expand": true,
      "tags": [
        "Autonomous Vehicles",
        "Pollution",
        "Renewable Energy",
        "Environmentalism"
      ],
      "related_proj": [
        153,
        9,
        5
      ]
    },
    {
      "student_name": "Hoshita U.",
      "mentor_name": "Ayush Pandit",
      "mentor_title": "PhD Candidate at Stanford, prior Stanford Bioengineering Alum",
      "mentor_image": "assets/images/mentor_imgs/8ebea060f7acf53ebda820482eaa4ee26d37df4b.png",
      "student_image": "assets/images/student_imgs/69691d7a6d2fb8cc071cd0ff3f9f6ca199544c90.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Using Linear Regression to Detect the Binding Efficiency of Ligands for Effective p53-MDM2 Inhibition",
      "project_desc": "Did you know that each day about 100-130 billion potentially cancerous cells undergo cell death? Every day, each person’s body prevents tumors from growing, and it’s all thanks to the p53 protein suppressor. Coded by the TP53 gene, the p53 tumor suppressor protein is responsible for signaling the cell repair and/or cell apoptosis pathway when damaged DNA growth is detected. In over 50% of human cells, the code for the p53 protein is corrupted, reducing its ability to inhibit tumor growth. When tumorous cells are present, there can be an overexpression of the MDM2 protein. The MDM2 protein is a protein that suppresses the activity of p53 making it even harder for controlling the potential development of cancer.\n\nThis research project targets the interaction between the MDM2 and p53 proteins to find out the most efficient ligands, or small molecules, that can bind to MDM2 and prevent the inhibition of p53 so as to stimulate the opportunity for p53 to signal for cell repair/death. This was accomplished by pre-processing a dataset of known MDM2-p53 inhibition ligands from chEMBL using Numpy/Pandas and adding necessary features into the dataset from rdKit. The processed data was then ran through a Sklearn regression model to accurately predict the efficiency of the ligands or the negative log of their Standard Value (binding efficiency).\n\nOverall, the model was able to perform quite well after optimization; however, it would produce even more accurate results if the images consisting of each ligand’s molecular structure could be processed as well since it would allow for a greater variety of features to draw patterns between. \n\nFor further research, this project will be expanded into predicting new ligands through given features based on the efficiency of known ligands with specific features. This, if accurate, would aid in discovering new ligands for cancer therapy and drug administration.",
      "research_paper": "assets/pdfs/faecfdcf79e0a492ed4e77e506b4f67e0b357f2f.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://embed.widencdn.net/img/criver/qmabxipjrb/640x426px/iStock-824641956.jpg?crop=true&keep=e&q=80&color=ffffffff&u=atq9fr",
      "headline": "This research project targets the interaction between the MDM2 and p53 proteins to find out the most efficient ligands, or small molecules, that can bind to MDM2 and prevent the inhibition of p53 so as to stimulate the opportunity for p53 to signal for cell repair/death.",
      "project_id": "b28a74bc5275cc2672fb36c1797ec49235a84a0f",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Cancer",
        "DNA"
      ],
      "related_proj": [
        58,
        10,
        119
      ]
    },
    {
      "student_name": "Ariel M.",
      "mentor_name": "Roger Jin",
      "mentor_title": "Computer Science MS at MIT, ML researcher at Apple",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": "Generating Instagram Captions with ViT-GPT2 and GPT3",
      "project_desc": "This paper presents a novel approach for generating Instagram captions based on visual features and language models. Our caption generator combines Vision Transformer-GPT2 and GPT3 to generate descriptive and engaging captions in the style of an Instagram post. Our caption generator has the potential to assist social media users, content creators, and marketers in generating high-quality and engaging captions for their Instagram posts. Future work includes exploring more advanced visual and textual features and incorporating user feedback and preferences.  Instagram is a popular social media platform for sharing photos and videos. One of the key elements of a successful Instagram post is a catchy caption that enhances the visual content and engages the audience. However, coming up with a good caption can be challenging, especially for users who are not native speakers of the language or lack creativity. In this paper, we propose an automatic caption generator that takes an image as input and generates an Instagram-style caption using state-of-the-art language models.",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://images.ctfassets.net/az3stxsro5h5/2w6SEmjPFbNKsDcGvKbhGN/14f0d987489db043a653e39f283b0b12/March8-10-Ways-to-Instantly-Improve-Your-Instagram-Bio-Horizontal.png?w=920&h=800&q=50&fm=png",
      "headline": "This paper presents a novel approach for generating Instagram captions based on visual features and language models. Our caption generator combines Vision Transformer-GPT2 and GPT3 to generate descriptive and engaging captions in the style of an Instagram post.",
      "project_id": "5f6afe37529fa4c059aa78e3f028f5a6736faa99",
      "expand": false,
      "tags": [
        "Social Media",
        "GPT-3",
        "Generative AI",
        "Pop Culture"
      ],
      "related_proj": [
        84,
        118,
        82
      ]
    },
    {
      "student_name": "Soham P.",
      "mentor_name": "Erick Ruiz",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics",
        "Engineering"
      ],
      "project_title": "Fake news detection, methods and data processing",
      "project_desc": "\tIn this project we studied several models to discern fake news articles from real news articles to find the best method. Fake news has become very prevalent in the digital age. There are thousands of articles that were written with the intent to mislead in various subjects: politics, social issues, and conspiracies. Leveraging two datasets that have true news articles and fake news articles we used 4 different machine learning models and compared them to each other to find the best method. We looked at the accuracy, the features, and the percentage of false positives the models gave. We found that the best model out of the four that we used was the custom neural network which got an accuracy of 95.7%. \n",
      "research_paper": "assets/pdfs/65a0dff3b3be40d8d58078fad6b029957de8ebae.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://static.vecteezy.com/system/resources/thumbnails/006/299/370/original/world-breaking-news-digital-earth-hud-rotating-globe-rotating-free-video.jpg",
      "headline": "In this project we studied several models to discern fake news articles from real news articles to find the best method.",
      "project_id": "6841a2ee4fd5265bb7aa9adecc64f4126b3eeb4e",
      "expand": false,
      "tags": [
        "Fake News",
        "News"
      ],
      "related_proj": [
        173,
        83,
        130
      ]
    },
    {
      "student_name": "Yash J.",
      "mentor_name": "Sarthak Kanodia",
      "mentor_title": "MS Data Science from Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/f4a63723d70977fd2076b085ca363a579ead7fac.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Generative Adversarial Fusion for Supercell Thunderstorm Forecasting",
      "project_desc": "Supercell thunderstorms have long evaded forecasters, leading to inadequate warning and preventable deaths. Current prediction methods use direct weather observation and algorithmic computer models, while they are not very computationally intensive, they lack in their ability to maintain accurate predictions.  Machine learning has been showing some promising results with generative networks proving to show the most promise. This paper proposes a novel machine learning model that attempts to address this problem by taking advantage of two different types of crucial data. My proposed architecture combines three distinct Machine Learning models: a Recurrent Deep Neural Network, a Wasserstein Generative Adversarial Network, and a Transformer Generative Adversarial Network to leverage numerical and visual data. While my research is purely theoretical at this stage, I believe that my proposed machine learning architecture has the potential to pave the way for future studies and practical implementations in the field of supercell thunderstorm prediction, providing significant improvements in forecast accuracy and lead time. \n",
      "research_paper": "assets/pdfs/877fbe0b4244ae2ebda304f7ba7bd6234477cc90.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "graphic_link": "https://media-cldnry.s-nbcnews.com/image/upload/t_fit-1500w,f_auto,q_auto:best/newscms/2014_21/452681/140519-science-cell.jpg",
      "headline": "This paper proposes a novel machine learning model that attempts to address this problem by taking advantage of two different types of crucial data.",
      "project_id": "028bbc5a631b1d7076ba0f607e05ea4edc80a5f4",
      "expand": false,
      "tags": [
        "Environmentalism",
        "Weather"
      ],
      "related_proj": [
        113,
        121,
        41
      ]
    },
    {
      "student_name": "Anya N.",
      "mentor_name": "Eric Bradford",
      "mentor_title": "Electrical Engineering and Computer Science Masters from MIT, Technical PM at Apple",
      "mentor_image": "assets/images/mentor_imgs/a1fe850d3c824ed0e95d61c0d97ce81721602779.png",
      "student_image": "assets/images/student_imgs/dfdaa723b50328e607df72570d675b7ab3ae8a3a.png",
      "domains": [
        "Sports and Motion Sciences"
      ],
      "project_title": "Revolutionizing Football: Using Machine Learning to Predict Future Performances for Quarterbacks",
      "project_desc": "Football is a largely popular American sport, and within fans of the sport, betting and fantasy football are also very popular. Football is also very unpredictable. On a certain day, a player may perform better because of the weather. On another day, that player may perform poorly because of an injury. Similarly, yearly stats also fluctuate. A quarterback may throw for 5,000 yards one year, and the next year they may throw for 3,000 yards. Due to this, it is important to have a reliable application that can aid users of betting and fantasy football in which players they should put bet on, or choose for their fantasy teams. My motivation behind this project was to create something that could further betting and fantasy football, and even increase traction. My approach to doing this was putting together dataframes of yearly statistics from 75 quarterbacks, and seeing if my linear regression model could predict statistics for a future season. Reliable sources such as the NFL website, or ESPN, were used in order to put these statistics together. The main objective was to produce as accurate as possible stats, which came over time as more seasons were added, and the code was tweaked. ",
      "research_paper": "assets/pdfs/f311dfcff79f25a124c87cb4a1cb4c7664f906ce.pdf",
      "project_yr": "2023",
      "project_quarter": "Winter",
      "project_webpage": "https://predictfootball.streamlit.app/",
      "graphic_link": "https://static.www.nfl.com/image/private/t_landscape_tablet/league/zlec4lzerzuibhd4e8og.jpg",
      "headline": "It is important to have a reliable application that can aid users of betting and fantasy football in which players they should put bet on, or choose for their fantasy teams. My motivation behind this project was to create something that could further betting and fantasy football, and even increase traction.",
      "project_id": "83b8dba6998311289d5b09fe6cd9939404c278f9",
      "expand": true,
      "tags": [
        "Sports",
        "Pop Culture"
      ],
      "related_proj": [
        11,
        142,
        49
      ]
    },
    {
      "student_name": "Thomas G.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/87c56ab56e04ac1e77c91a72265ff270fb70c4c1.png",
      "domains": [
        "Environment"
      ],
      "project_title": "SoilingNet: Convolutional Neural Networks for Analysis of Soiling Defects in Photovoltaic Panels",
      "project_desc": "Solar panel soiling is an important problem for the energy sector as we transition\naway from fossil fuels. Previous methods have aimed to analyze and combat soiling\nthrough the use of classical computer vision techniques or weakly supervised deep\nlearning. In our study, we present a two-step, fully supervised deep learning approach\nfor analyzing solar panel soiling on a per-panel level. From a single RGB image of a\nsolar panel, combined with the environmental factor of solar irradiance, our method\nproduces predictions for soiling type, soiling location, and soiling severity. We also\nintroduce a first-of-its-kind dataset labeled with ground truth semantic segmentation\nmaps for the purpose of solar panel soiling analysis. This dataset contains 1104\nsamples. We find that our model can achieve Jaccard indices in excess of 70 when\npredicting semantic segmentation maps and top-1 accuracy’s in excess of 92% when\npredicting soiling severity. Our method outperforms previous techniques in the\ndomain of solar panel soiling analysis.",
      "research_paper": "assets/pdfs/b029cffc6a739c9459cc16459749d67b5e1b03e9.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "github": "https://github.com/ThomasGust/SoilingNet",
      "graphic_link": "https://cdn.greenmatch.co.uk/cdn-cgi/image/format=auto/2/2023/08/iStock-1302801693-1-1024x683.jpg",
      "headline": "In our study, we present a two-step, fully supervised deep learning approach for analyzing solar panel soiling on a per-panel level.",
      "project_id": "c222e4830d9eece578ea0265e8615520340610ed",
      "expand": true,
      "tags": [
        "Environmentalism",
        "Solar Power",
        "Renewable Energy"
      ],
      "related_proj": [
        70,
        85,
        107
      ]
    },
    {
      "student_name": "Michael Z.",
      "mentor_name": "Kasra Koushan",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/d3fd4bd30ab402141be3f6ed665975a9bbc84492.png",
      "domains": [
        "PoliSci and Law"
      ],
      "project_title": "Predicting Populations: Modeling Demographic Predictions for Nations Around the World Using Population Pyramids and Demographic Transition Models",
      "project_desc": "Examining and analyzing demographic statistics are a key foundation for analyzing the makeup of a population for a given nation. Some of the most important population demographics include total population, crude birth rate(CBR - births/1000 people), crude death rate(CDR - deaths/1000 people), and populations of different age groups and genders. These help provide a basis into understanding the overall population growth and stage of development of a particular country, which can ultimately be used to shape economic, political, and social policies for different nations around the world. In this research project, I used multiple machine learning models(neural network and linear regression) in order to predict key demographic statistics over the next 5 years for each nation. These models yielded accurate results, as most sets of predictions had accuracy scores above 0.85. Additionally, using advanced visualization techniques, I modeled these predictions using popular human geography models such as population pyramids and demographic transition models. I found that for more developed nations such as countries in Western Europe and East Asia, the population of older individuals is going to rise significantly over the coming years. For less developed nations such as those in Africa, the overall population is going to skyrocket, and the proportion of young adults who will enter the labor force will significantly surge. In terms of real-world implications, these results could lead to more developed nations starting to spend more on social security and less developed nations starting to implement policies that limit population growth.",
      "research_paper": "assets/pdfs/4700bbd9cc1be659e07eb2654d65a8f2513ea5f1.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://i0.wp.com/www.sciencenews.org/wp-content/uploads/2022/11/111422_ag_8-billion-people_feat.jpg?fit=1030%2C580&ssl=1",
      "headline": "In this research project, I used multiple machine learning models(neural network and linear regression) in order to predict key demographic statistics over the next 5 years for each nation.",
      "project_id": "c7ed18c5ec1b1ab442b32c81db290943de96e4cf",
      "expand": true,
      "tags": [
        "Sociology",
        "Political Science"
      ],
      "related_proj": [
        165,
        79,
        78
      ]
    },
    {
      "student_name": "Ethan V.",
      "mentor_name": "Anjali Singh",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/99206a7ff5b50a947138a2629eadbf6bf85c3aaf.png",
      "domains": [
        "Engineering",
        "Healthcare and Biology"
      ],
      "project_title": "Improving User Retention in Video Game Industry",
      "project_desc": "This research paper delves into the intersection of player data analytics and affective computing to predict and adapt game difficulty levels for the purpose of enhancing player retention in the gaming industry. The study used two Kaggle datasets which include a facial emotion dataset and real-time call of duty player data. A novel emotion detection model, leveraging transfer learning, is constructed to discern a spectrum of emotions from player facial expressions. The emotional context extracted from this model is seamlessly integrated with player data, facilitating a supervised learning task centered on predicting changes in game difficulty levels—a pivotal factor in player engagement. Several machine learning models including K-Nearest Neighbors, Decision Trees, Random Forests, MLP Classifiers, AdaBoost, and Gaussian Naive Bayes were employed to use emotion and player data to predict difficulty changes that will raise player retention.",
      "research_paper": "assets/pdfs/381bc6b694f289f684e8fd9eb46c960e5cd954bd.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://media.cnn.com/api/v1/images/stellar/prod/i-stock-1287493837-1.jpg?c=16x9&q=w_800,c_fill",
      "headline": "This research paper delves into the intersection of player data analytics and affective computing to predict and adapt game difficulty levels for the purpose of enhancing player retention in the gaming industry.",
      "project_id": "61b6744c6bb550e4068d693e1c0ab942d2751c91",
      "expand": false,
      "tags": [
        "Video Games",
        "Product Analysis"
      ],
      "related_proj": [
        173,
        59,
        157
      ]
    },
    {
      "student_name": "Will K.",
      "mentor_name": "Ana Sofia Muñoz Valadez",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/9ef2e63ea5415702f7abca28b725ad7cd5f59e16.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Using Neural Networks to Predict U.S. Corporate Profits on Electronic Goods",
      "project_desc": "The 2018 trade war between the United States and China, COVID-19, supply chain disruptions, component costs, and post-COVID consumer behavior have all created negative impacts on U.S. corporate profits on electronic goods. The goal of this project is to train two neural network AI models: a Multi-Layer Perceptron (MLP) neural network and a Long Short-Term Memory (LSTM) neural network, to predict U.S. corporate profits on electronic goods into the future. We introduce 1) the datasets used for our models, 2) describe our MLP neural network AI model and its corresponding results, 3) describe our LSTM neural network AI model and its corresponding results, and finally 4) summarize our findings and offer suggestions for future research into this area of study. Our hypothesis for the project was that even though there was a boost to U.S. corporate profits after COVID-19 due to re-opening of the economy and through government aid, these gains were completely offset by the much stronger factor of supply chain inefficiencies, thus bringing these profits down.  Our results show that supply chain inefficiencies have begun to abate worldwide and as they do, growth in corporate profits continues to stabilize to pre-pandemic levels.",
      "research_paper": "assets/pdfs/d33f8d1719612e5683febd93d6b3b6b7ed1cbf73.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "github": "https://github.com/WilliamKrofchik/InSpiritAI-Neural-Networks-US-Corporate-Profits",
      "graphic_link": "https://www.ecommercetimes.com/wp-content/uploads/sites/5/2023/12/ecommerce-success.jpg",
      "headline": "The goal of this project is to train two neural network AI models: a Multi-Layer Perceptron (MLP) neural network and a Long Short-Term Memory (LSTM) neural network, to predict U.S. corporate profits on electronic goods into the future.",
      "project_id": "4b66e513746ea2fc104cdc9cf98f036d1ce77873",
      "expand": false,
      "tags": [
        "E-Commerce",
        "Consumer Behavior"
      ],
      "related_proj": [
        2,
        88,
        122
      ]
    },
    {
      "student_name": "Neil S.",
      "mentor_name": "Ye Wang",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Optimizing Stroke Detection with Machine Learning Techniques",
      "project_desc": "          This paper addresses the critical question of how various medical factors influence vulnerability to stroke in patients, aiming to develop an effective machine-learning model for stroke detection. Stroke remains a pervasive health concern, with one in six deaths from cardiovascular disease attributed to it in 2021. Every 40 seconds in the United States, an individual experiences a stroke, emphasizing the urgent need for accurate prediction methods. Our study leveraged diverse machine learning models, necessitating dataset balancing prior to model evaluation. Subsequently, a comparison of metrics and fine-tuning of hyperparameters culminated in the selection of the K-Nearest Neighbors (KNN) as the optimal model. The KNN exhibited notable performance metrics, achieving an accuracy score of 80%, a recall score of 89.2%, an F1 score of 83.3%, and a precision score of 78.1%. These metrics signify the model's proficiency in correctly identifying stroke cases, its sensitivity to true positives, and its balance between precision and recall. This study underscores the potential of advanced machine learning models, particularly the KNN, in stroke detection, suggesting its viability for integration into medical consultations. However, the gravity of stroke as a life-threatening condition necessitates rigorous testing on larger and more recent datasets to ensure robustness and reliability in clinical applications. It should also be noted that changing the BMI, average glucose level, and age of the patient affected model performance the most.",
      "research_paper": "assets/pdfs/c17c2a2aacd0ad37c9317ed01b906dc4e0db52c4.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://www.cnet.com/a/img/resize/e84274d0e20b455830c300382bf63baec528b594/hub/2015/01/22/ecb2ee46-95f1-459d-86a0-e47754031a58/edsap-samsung-stroke-system.jpg?auto=webp&width=1200",
      "headline": "This paper addresses the critical question of how various medical factors influence vulnerability to stroke in patients, aiming to develop an effective machine-learning model for stroke detection.",
      "project_id": "d039e4651663265c2b5b5254617d09b10bebba4b",
      "expand": false,
      "tags": [
        "Strokes",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        125,
        59,
        12
      ]
    },
    {
      "student_name": "Vanessa H.",
      "mentor_name": "Matan Gans",
      "mentor_title": "Computer Science MS at Brown",
      "mentor_image": "assets/images/mentor_imgs/73681a9d5dfa8c731c6630f956bd88bbcea828bc.png",
      "student_image": "assets/images/student_imgs/07e9e0ae4c6a0930d5dc359b9e69b119cd2a21a3.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": "Sign Language Recognition In Deep Learning: A Comparative Study of Custom CNN Model and Pre-Trained Architectures",
      "project_desc": "The goal of this research is to successfully train\ndifferent Convolutional Neural Network (CNN) models to\nidentify sign language images, compare the performance of\neach model, and figure out the best model for image\nrecognition and classification. The dataset is collected from\na public domain called Kaggle and contains 2515 images\nrepresenting English alphabet letters and digits from 0-9 in\nAmerican Sign Language. We use three different deep\nlearning architectures to approach this computer vision\nproblem. Specifically, we implement a custom CNN model\nand fine-tune two pre-trained models called InceptionV3 and\nResNet50. These models were trained using TensorFlow and\nKeras. Each model consists of numerous convolutional and\nfully connected layers and is trained to classify images into\none of 36 different classes. The models' architecture\nfacilitates feature extraction and classification tasks, making\nit well-suited for image recognition problems. The\nperformance of the custom CNN model, InceptionV3, and\nResNet50 are 94.17%, 96.68%, and 98.67%. All three\nmodels generate extremely high accuracy which can\ncontribute to helping researchers choose the best\ndeep-learning method for Sign Language recognition and\nother image classification tasks.",
      "research_paper": "assets/pdfs/45ecc38f655fcdb6d9281ccc6541750a3170da4c.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://techcrunch.com/wp-content/uploads/2019/08/handtracking_shot.png",
      "headline": "The goal of this research is to successfully train different Convolutional Neural Network (CNN) models to identify sign language images, compare the performance of each model, and figure out the best model for image recognition and classification.",
      "project_id": "6f78e24f676fcfdce1480095509f10b44837dd52",
      "expand": true,
      "tags": [
        "Natural Language Processing",
        "Modern Languages",
        "CNNs"
      ],
      "related_proj": [
        51,
        162,
        79
      ]
    },
    {
      "student_name": "Daniel S.",
      "mentor_name": "Jonathan Delgadillo Lorenzo",
      "mentor_title": "Mathematical Economics UPenn Alum, Harvard Humanities Researcher, Data Scientist at Theta",
      "mentor_image": "assets/images/mentor_imgs/21734ecfe074515561989e3cc073df8fc46aee21.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Height Prediction Using Basic Data ",
      "project_desc": "With the use of an accurate height prediction, doctors understand a patient better by seeing if they are under or above their predicted height. I used basic data sets to see if some learning models have a chance of predicting height based on country and age. Decision tree regressor, Linear regression, and random forest regressions were successful models that I used. The AI prediction varied between a miss of 40cm all the way to 100 cm throughout all models. But After hyper tunneling the model, it got to a closely stable average of 17 cm off the real height. Although being off 17 cm is not respectable in the Medical World, this may prove that with higher ends of data, height prediction may be applicable to medical use. ",
      "research_paper": "assets/pdfs/631c5d450a46c4a4681f50396896ed670d6b0b2e.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://pyxis.nymag.com/v1/imgs/e64/f95/03ff31bb09aaa19303219fcbb7b5e34abf-20-comparative-height.jpg",
      "headline": "I used basic data sets to see if some learning models have a chance of predicting height based on country and age.",
      "project_id": "8838703db15c71a335ea938b1fb380ab03726bcf",
      "expand": false,
      "tags": [
        "Physiology",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        42,
        157,
        163
      ]
    },
    {
      "student_name": "Alya K.",
      "mentor_name": "West Foster",
      "mentor_title": "PhD Candidate in Electrical Engineering at Stanford",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/570bdaf2ea1de7da66cab731ba42b44aef4a7053.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Examination of Dragonflies (Pantala Flavescens): Characteristics, Identification and Migration Paths",
      "project_desc": "The Pantala flavescens is a long range migratory dragonfly known for its widespread global population. Some recent research has suggested it can travel 4500 km in its lifetime, making it one of the farthest known migratory insects. This study explores the Pantala flavescens species' anatomy and mechanisms for long-distance migrations. Convolutional neural networks and transfer learning were used for image recognition of Pantala flavescens. Stable hydrogen isotope values are analyzed to locate the species' geographic origin. By gaining a better understanding of the Pantala flavescens’ definite migration routes, the study aims to minimize potential harm from climate change.",
      "research_paper": "assets/pdfs/7d93ce84310b6eabbbf62d202cf6adf538907da4.pdf",
      "project_yr": "2022",
      "project_quarter": "Winter",
      "graphic_link": "https://www.treehugger.com/thmb/1UGEjB6ykn52cVpLFitfw2vqwsY=/1500x0/filters:no_upscale():max_bytes(150000):strip_icc()/4-bc3246043abe473ab750f039b5c70b53.jpg",
      "headline": "This study explores the Pantala flavescens species' anatomy and mechanisms for long-distance migrations.",
      "project_id": "f8bd2c06742c4deb8185ee4142035293796c3d51",
      "expand": true,
      "tags": [
        "Wildlife"
      ],
      "related_proj": [
        68,
        27,
        107
      ]
    },
    {
      "student_name": "Sahir G.",
      "mentor_name": "Landon Butler",
      "mentor_title": "PhD Candidate at Berkeley, UPenn CS Master's",
      "mentor_image": "assets/images/mentor_imgs/d555e67db872a4dce56c1938f2efc953bb18ce3f.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Business and Finance",
        "Environment"
      ],
      "project_title": "Predicting International Commodity Prices, Production, and Land Usage Towards Reducing Agricultural Emissions ",
      "project_desc": "Our long-term goal for this research project was to analyze food insecurity and determine if we could predict the emissions and pricing of different food commodities. Towards this goal, we needed to understand the changes in food production, land use, efficiency, and pricing of these commodities in a sample of different countries. Given the rampant food insecurity heavily influenced by climate change, aiding in reducing the impacts of climate change could be a crucial step towards alleviating food insecurity. Our approach to answering this question was to first gather and analyze data and then use an univariate ARIMA model for time series to further develop our hypothesis and figure out if it was possible to predict future trends. Despite not including external factors like world conflicts, inflations, weather, and trade, the univariate ARIMA is able to predict the prices, production, and efficiency well on some commodities for some countries. Our major conclusions were that our research could be used to make cursory predictions for future values of the food prices and agricultural emissions. ",
      "research_paper": "assets/pdfs/4be847804b26e23f7e146d24b58e3a021c1bc2fe.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://s3.eu-west-2.amazonaws.com/iema.net/transform/_720x520_crop_center-center_80_none/154422/web_p5_agricultural-emissions_istock-1201461261.webp",
      "headline": "Our long-term goal for this research project was to analyze food insecurity and determine if we could predict the emissions and pricing of different food commodities. ",
      "project_id": "229d998bb0c3ad8c33a6edec3a7e0c7544463dbc",
      "expand": true,
      "tags": [
        "Real Estate",
        "Pricing",
        "Farming"
      ],
      "related_proj": [
        125,
        123,
        147
      ]
    },
    {
      "student_name": "James S.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/39d121d6f1ed36ac5070c213bfa0fbf4bed6e926.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Unmasking Fraud: Applying Machine Learning to Detect Bank Drops",
      "project_desc": "Bank drops are fraudulent bank accounts that are used to funnel and transfer illegal funds. Most fraudulent bank accounts often go undetected, exposing flaws in the banking system and require large teams of cybersecurity specialists to spend significant amounts of time and money trying to correct. This paper evaluates eight machine learning (ML) models that are capable of flagging bank accounts as fraudulent at the time of bank account registration. Each of these models process bank account registration information to instantly determine the legitimate versus fraudulent status of an account, expediting the bank account flagging process, saving banks internal resources and protecting bank clients from fraudulent activities. The highest performing models are ensemble models such as the gradient boosting classifier and the random forest classifier, each having accuracy scores above 90%. These results show that ML models are a valuable tool to detect fraudulent accounts and protect banks across the world, effectively and efficiently stopping crime before it begins.",
      "research_paper": "assets/pdfs/4ea1bd90e4a3b93d3a07fe2f040afaa1f8fd729e.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://static6.depositphotos.com/1007113/629/i/450/depositphotos_6292715-stock-photo-imposing-bank.jpg",
      "headline": "This paper evaluates eight machine learning (ML) models that are capable of flagging bank accounts as fraudulent at the time of bank account registration.",
      "project_id": "603bda285bb9d365b98a38e2314c59751886eab0",
      "expand": false,
      "tags": [
        "Fraud"
      ],
      "related_proj": [
        147,
        116,
        186
      ]
    },
    {
      "student_name": "Gregory G.",
      "mentor_name": "Ribhav Gupta",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/3cb1d6b771d158ca05ed87c9ec59fe1d0d2ba45e.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Predicting State-Wide Cotton Yield using Geospatial Data",
      "project_desc": "Forecasting cotton yield has immense interest in the agriculture world considering how valued the crop is world wide. However, current methods of predicting cotton yield remain significantly unreliable since they rely on human assumption. Additionally, these forecasts are significantly impacted by climate change. This paper aimed to use spatial models to better predict cotton yield. Prior papers on predicting cotton production have yielded fair results utilizing non-spatial models [1]. Therefore, the focus of this study was to use Geospatial Linear Regression to predict cotton yield. Through this study, it was found that using climate, temporal, and geographic data cohesively produces the most accurate predictions. Additionally, the Random Forest model, with an MSE of 2.67, performed better than the Geospatial model, with an MSE of 8.22. This concludes that non-spatial models are better for predicting state-wide cotton yield when compared to spatial models. This research provides valuable results to the agriculture setting as well as different aspects to further this research. ",
      "research_paper": "assets/pdfs/8d281116815e2ca445afd4d6d4bf9cf346461864.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://upload.wikimedia.org/wikipedia/commons/2/22/Cotton_field_kv17.jpg",
      "headline": "The focus of this study was to use Geospatial Linear Regression to predict cotton yield.",
      "project_id": "c563b1d8921cad372ca0c933b012be3fc9910082",
      "expand": false,
      "tags": [
        "Earth Systems",
        "Farming"
      ],
      "related_proj": [
        27,
        145,
        85
      ]
    },
    {
      "student_name": "Alfred B.",
      "mentor_name": "Ayush Pandit",
      "mentor_title": "PhD Candidate at Stanford, prior Stanford Bioengineering Alum",
      "mentor_image": "assets/images/mentor_imgs/8ebea060f7acf53ebda820482eaa4ee26d37df4b.png",
      "student_image": "assets/images/student_imgs/63561bbf9e689de55369fb1dc997a0613f279775.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Multi-Label Prediction of Protein Subcellular Localizations through Machine Learning and State-of-the-art Structural Embeddings",
      "project_desc": "A polypeptide, more commonly referred to as a protein, is a macromolecule consisting of linear sequences of organic compounds called amino acids joined together by covalent bonds. Proteins carry different functions in living organisms, which are often associated with their location inside eukaryotic cells. Experimental approaches for localizing proteins such as green fluorescent protein and microscopic detection are costly and time-consuming. Thus, computational methods have been adopted to facilitate this process by generating accurate predictions. This research explores various machine learning models to predict protein subcellular localization using UniProt’s Swiss-Prot database and Meta’s high-performing ESM2 structural embeddings. We extracted our dataset’s protein embeddings from ESM2’s model code publicly posted on GitHub. Multiclassification Scikit-learn models were trained and tested to achieve optimal predictions. Our best-performing model, a weighted 4-Nearest Neighbors Classifier, yielded a 57% accuracy and a MacroF1 score of 66%. Our classifier outperformed past state-of-the-art algorithms (such as Deeploc2.0) on several but not all metrics. The results obtained demonstrate the potential of machine learning in producing unprecedented advancements in the proteomics field.",
      "research_paper": "assets/pdfs/b4605e9f73e6265e396866ee541a2dc83e4834b6.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "graphic_link": "https://www.researchgate.net/publication/343240378/figure/fig1/AS:928762371788803@1598445613852/MlPar-6-protein-subcellular-localization-during-the-early-development-of-M-leidyi.png",
      "headline": "This research explores various machine learning models to predict protein subcellular localization using UniProt’s Swiss-Prot database and Meta’s high-performing ESM2 structural embeddings. ",
      "project_id": "7db277df6e82f85fb5943f78ce202129619ee44f",
      "expand": true,
      "tags": [
        "Biology",
        "DNA"
      ],
      "related_proj": [
        183,
        131,
        101
      ]
    },
    {
      "student_name": "Stanley C.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Environment",
        "Healthcare and Biology"
      ],
      "project_title": "Developing a novel 3D GNN and Random Forest Regression model for screening and predicting potential oxide electrocatalysts with greater accuracy and computational efficiency",
      "project_desc": "In order to develop green hydrogen into a viable source of renewable energy that can be produced and stored at scale to address climate change, cost effective electrocatalysts (catalysts that facilitate the reaction of electrolysis), are imperative. Current simulation based efforts, such as Density Functional Theory (DFT), to screen such catalysts are too computational intensive and not scalable. To address this, ML models have been developed that predict the total relaxed energy of catalyst structures, a key attribute that drives chemical reactions. Phase 1 of my research evaluated the Dimenet++ model and Graphormer 3D Transformer model on a subset of the OC20 (Open Catalyst 2020) IS2RE dataset to analyze the relationship between a 3D Transformer and a GNN, and establish a baseline model that can be used to compare to Phase 2. In addition, I trained a joint model with both OC20 and OC22 data and compared the results to the independently trained models produced earlier to evaluate joint training on model performance. However, the thermodynamic data provided by OC20 is not enough for accurate screening. Atom charge distribution is an important factor in driving the chemical reactions of electrocatalysts so it can potentially also bolster model performance, and it has not yet been included in previous models because it is calculated through DFT. Using Bader Charges calculated through DFT for atom structures in OC20, I performed feature engineering with 13 additional atomic characteristics, and constructed a Random Forest and GNN model to predict Bader charges for all atom structures. I then modified the Graphormer model to incorporate this new predicted charge attribute. The refined model saw ~3.5% less in loss and ~10% greater in EwT across all the validation splits. With this new model, electrocatalysts can be screened faster with higher accuracy at scale.",
      "research_paper": "assets/pdfs/cb5813a6a122d09c165039fb28c40dede1d0739c.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://u4d2z7k9.rocketcdn.me/wp-content/uploads/2023/07/Untitled-683-%C3%97-1024px-1024-%C3%97-683px-2023-07-12T154605.688.jpg",
      "headline": "Phase 1 of my research evaluated the Dimenet++ model and Graphormer 3D Transformer model on a subset of the OC20 (Open Catalyst 2020) IS2RE dataset to analyze the relationship between a 3D Transformer and a GNN, and establish a baseline model that can be used to compare to Phase 2.",
      "project_id": "12d0ba551ac91ec981010b287a4026f75ca451bf",
      "expand": true,
      "tags": [
        "Renewable Energy",
        "Earth Systems"
      ],
      "related_proj": [
        15,
        41,
        115
      ]
    },
    {
      "student_name": "Ashwith Y.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Predicting Mental Health Conditions Using Student Demographic Information ",
      "project_desc": "We are working with supervised data and this is classification data. With this, we will train the data to extrapolate what symptoms the user has based on their demographics and academic/social life. This can help understand what symptoms a person has of certain characteristics. Age, gender, course year, course study, cgpa, and marital status are the inputs of the data. Whether the person is more prone to depression, anxiety, or panic attacks is the output of the model.",
      "research_paper": "assets/pdfs/517cc29d68fdde1c6a7213015a94959c85956b74.pdf",
      "project_yr": "2024",
      "project_quarter": "Winter",
      "graphic_link": "https://onlinecounselingprograms.com/wp-content/uploads/sites/52/2021/04/9646_OCP_Managing-Your-Mental-Health-in-College-hero.jpeg",
      "headline": "With this, we will train the data to extrapolate what symptoms the user has based on their demographics and academic/social life. This can help understand what symptoms a person has of certain characteristics.",
      "project_id": "4b4af9496d29d87deae52a285bf64b58f22b0c39",
      "expand": false,
      "tags": [
        "Mental Health",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        58,
        10,
        7
      ]
    },
    {
      "student_name": "Taneesh S.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/402e61f34449089e9ff9432d6f8a67bd30b6dc7e.png",
      "domains": [
        "Psychology"
      ],
      "project_title": "Utilizing Artificial Intelligence for the Identification of Students with Depression and Anxiety through Social Media Analysis",
      "project_desc": "In response to the escalating concerns about the mental well-being of students, this research represents a significant stride in utilizing artificial intelligence (AI) to discern individuals experiencing depression and anxiety through their social media comments. The study employs a comprehensive suite of regression methods, including Logistic Regression, Decision Trees, Support Vector Classifier (SVC), Random Forest, and the Ridge model. The dataset, meticulously curated with 6982 comments sourced from Kaggle, undergoes processing, incorporating a conscientious split into training and testing sets for robust evaluation. The AI models exhibit remarkable performance metrics, with precision reaching an impressive 99%, F1 scores achieving 99% for normal and 95% for depressive comments, and accuracy and weighted averages standing at 99%. The study serves its immediate purpose of identifying students on social media and hints at broader applications across diverse contexts. Beyond specific demographics, the research underscores the potential of AI in addressing mental health challenges, offering insights that extend far beyond the confines of student populations, thereby contributing to the broader discourse of AI's impact on mental health.",
      "research_paper": "assets/pdfs/70503acc5f63489fd8c4465b7c5a608133291e89.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://insights.som.yale.edu/sites/default/files/styles/max_2600x2600/public/2021-06/210606_AddictiveSocialMedia_FINAL_0.png?itok=Eiw5yNzF",
      "headline": "In response to the escalating concerns about the mental well-being of students, this research represents a significant stride in utilizing artificial intelligence (AI) to discern individuals experiencing depression and anxiety through their social media comments.",
      "project_id": "a4ebfa3dbea353010959ac3d581a4955f0506ef4",
      "expand": true,
      "tags": [
        "Mental Health",
        "Social Media"
      ],
      "related_proj": [
        71,
        58,
        61
      ]
    },
    {
      "student_name": "Siddharth M.",
      "mentor_name": "Anthony Cuturrufo",
      "mentor_title": "PhD Candidate in Computer Science at UCLA, Cornell CS Alum",
      "mentor_image": "assets/images/mentor_imgs/f699a372eb08ca69ee86a5edf71437a7c68bccd5.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Arts and Music"
      ],
      "project_title": "Classifying AI-Generated Music with AI Models ",
      "project_desc": "The emergence of artificial intelligence (AI) in music composition has surfaced as a contentious yet captivating topic of exploration within the music industry [1]. It raises questions about the legitimacy of machine-generated creativity and the complexities surrounding copyright, ownership, and the evolving nature of artistic expression [2]. One prominent and groundbreaking creation that has captured the attention of many is Jukebox, the AI music generation model created by OpenAI [3]. The importance of this research is twofold: addressing copyright issues and fostering the adoption of AI in the music industry. As AI-generated music blurs the line between human and AI creation, concerns regarding copyright ownership and artistic attribution become paramount [2]. To address these challenges, we created an AI model to differentiate AI-generated music from human-composed music, specifically focusing on music generated by the JukeBox model. Using 3 different models: a Multi-Layer-Perceptron (MLP), a long-short term memory network (LSTM), and a transformer, we were able to classify JukeBox generated songs with 96 percent accuracy.\n",
      "research_paper": "assets/pdfs/2ededfe6e527c0b86f86b57d56647c1400ba9a27.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "github": "https://github.com/0sid0/AI-Generated-Music-Detection.git",
      "graphic_link": "https://kentwired.com/wp-content/uploads/2023/08/AIPhotoAlton-1200x873.png",
      "headline": "The importance of this research is twofold: addressing copyright issues and fostering the adoption of AI in the music industry. As AI-generated music blurs the line between human and AI creation, concerns regarding copyright ownership and artistic attribution become paramount [2]. To address these challenges, we created an AI model to differentiate AI-generated music from human-composed music, specifically focusing on music generated by the JukeBox model.",
      "project_id": "0d221e1c32a9166edee886d980d9fefb06705a18",
      "expand": false,
      "tags": [
        "Music",
        "Generative AI",
        "Classification"
      ],
      "related_proj": [
        176,
        102,
        144
      ]
    },
    {
      "student_name": "Maanya M.",
      "mentor_name": "Anthony Cuturrufo",
      "mentor_title": "PhD Candidate in Computer Science at UCLA, Cornell CS Alum",
      "mentor_image": "assets/images/mentor_imgs/f699a372eb08ca69ee86a5edf71437a7c68bccd5.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Classifying Non-Gaussian Transient Noise in LIGO ",
      "project_desc": "Space exploration methods have never been as diverse and thorough as it is in today’s time. New techniques that span far beyond the conventional method of using electromagnetic radiation to identify and study astronomical evidence have been discovered and are now amplified using machine learning technology.  One such technique is using interferometers to record gravitational waves released from astronomical events.  Using computer vision to analyse gravitational waves can provide valuable information about previously obscured information. Also, gravitational waves are fundamentally unrelated to electromagnetic radiation thereby providing us with information that is entirely different from what we previously could access. This has consequentially opened an altogether new avenue for research and further exploration. The contrastive model for machine learning-based detection and classification performed the best. \n",
      "research_paper": "assets/pdfs/2f95d10505a390cda284941ed27390c4caebee52.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://community.cadence.com/resized-image/__size/1280x960/__key/communityserver-discussions-components-files/38/pastedimage1665668018635v1.png",
      "headline": "Using computer vision to analyse gravitational waves can provide valuable information about previously obscured information.",
      "project_id": "2aee5ef4fe79ef6b70c97e4b1f665db4ea5c4225",
      "expand": false,
      "tags": [
        "Physics",
        "Astronomy"
      ],
      "related_proj": [
        132,
        91,
        25
      ]
    },
    {
      "student_name": "Zachary R.",
      "mentor_name": "Zachary Gittelman",
      "mentor_title": "MS in Information Systems from Cornell; startup co-founder",
      "mentor_image": "assets/images/mentor_imgs/b80905554784ba740eb49afb192e8f2b990e2ad0.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Predicting the Performance of Code using Machine Learning",
      "project_desc": "The goal of this research paper is to see the potential of machine learning to predict the performance of a given program. Doing this could allow for faster predictions than compiling and properly testing a given program, especially with high compile-time optimization, allowing for faster optimization of a program. To accomplish this, we trained a model consisting of a TreeLSTM and an MLP on a dataset sourced from CodeChef in order to predict the time and memory taken by a given program. After training, we received an average percent error of 22.43% and 1261.07% for time and memory respectively which is fairly poor overall but somewhat fine for the time. As such, this paper concludes that machine learning can potentially be used to accomplish this goal but would likely need a more involved and thought through implementation to work effectively.",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://imageio.forbes.com/blogs-images/forbestechcouncil/files/2019/01/canva-photo-editor-8-7.png?height=640&width=640&fit=bounds",
      "headline": "The goal of this research paper is to see the potential of machine learning to predict the performance of a given program.",
      "project_id": "dec39b5fc78709d9127bd63ae43cf80e085e1853",
      "expand": false,
      "tags": [
        "Engineering",
        "Optimization"
      ],
      "related_proj": [
        180,
        150,
        30
      ]
    },
    {
      "student_name": "Smriti R.",
      "mentor_name": "Varsha Sandadi",
      "mentor_title": "MS at MIT, Computer Science MIT Alum",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/c7ba0b6967013acc4c64dc20a755f6188c845276.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Individualizing Medicine: AI in Predicting the Effect of Yoga on Adrenomedullin Levels",
      "project_desc": "Non-communicable diseases(NCDs) are the biggest cause of death globally, and have no steadfast vaccine or cure. However, preventative measures such as yoga have been found to prevent NCDs as well as mediate Adrenomedullin(ADM) levels, a biomarker for NCDs, in practitioners. While this relationship has been studied, I plan to study how yoga can aid an individual’s efforts to lower their ADM levels. I did this by creating and tuning four regression models that predicted a patient’s ADM levels after doing yoga: a Linear regressor, a Random Forest regressor, a Ridge regressor, and an Artificial Neural Network(ANN).  I found that the ANN had the smallest mean absolute error(MAE) and mean squared error(MSE) values, which meant that the ANN’s predicted values for ADM levels were closest to the actual values compared to the other models.\n",
      "research_paper": "assets/pdfs/d82bc016624c012f4ad361fc94cdf2d8d7b33f96.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://files.nccih.nih.gov/yoga-gettyimages-1204500395-16-9.jpg",
      "headline": "I created and tuned four regression models that predicted a patient’s ADM levels after doing yoga: a Linear regressor, a Random Forest regressor, a Ridge regressor, and an Artificial Neural Network(ANN). ",
      "project_id": "0675e2316aa4417f0bf8ac1567b3825acde3bad0",
      "expand": true,
      "tags": [
        "Physiology",
        "yoga"
      ],
      "related_proj": [
        98,
        55,
        29
      ]
    },
    {
      "student_name": "Aaron W.",
      "mentor_name": "Ronil Synghal",
      "mentor_title": "UPenn MS in Mechatronics and Robotics, Prior Full-Stack Software Engineer at Healthcare Startup",
      "mentor_image": "assets/images/mentor_imgs/b4d4da066f156a8ff549472bd1d2673cf8cde8e1.png",
      "student_image": "assets/images/student_imgs/a1d1c21a712b9810f91c00e5d47af215e3392350.png",
      "domains": [
        "Physics",
        "Engineering"
      ],
      "project_title": "Using Machine Learning Models to Analyze the Aerodynamic Properties of Airfoils",
      "project_desc": "In aerodynamics, the lift-drag ratio of an airfoil is of paramount importance, as it directly correlates with fuel efficiency. Traditionally, determining airfoil properties necessitates expensive experimental tests or computationally intensive simulations. This research posits that leveraging artificial intelligence could significantly reduce financial and computational costs while identifying optimal airfoil geometries. The data originates from Bigfoil, a database of airfoils compiled from UIUC’s database, Javafoil, and NACA tested airfoils. Specifically, we employed a convolutional neural network (CNN) to analyze various airfoil images and geometric data. The image data provides the machine learning system a visual understanding of the airfoil, and allows it to pinpoint aerodynamic properties to different shapes. Our model, constructed with multiple Conv2D layers and auxiliary components, was designed to predict the lift-to-drag ratio. Preliminary results were promising: the optimal CNN configuration accurately predicted the ratios for 58.6% of previously unseen airfoils. However, there remains potential for improvement. Enhanced hyperparameter tuning could further augment the prediction accuracy. Additionally, utilizing higher-resolution images and improving image quality enhance the network's predictive capabilities.",
      "research_paper": "assets/pdfs/d17fc8b72ba70f466e26e1ca5c504e79d3af959c.pdf",
      "project_yr": "2022",
      "project_quarter": "Summer",
      "github": "https://github.com/aron-the-great/AirfoilML",
      "graphic_link": "https://media.wired.com/photos/5b3ac9899a7504731f8818f8/master/pass/Quiet-NASA-Transpo.jpg",
      "headline": "This research posits that leveraging artificial intelligence could significantly reduce financial and computational costs while identifying optimal airfoil geometries.",
      "project_id": "9bfabfb410728099cea1e6eb1a49eeae30540f25",
      "expand": true,
      "tags": [
        "Engineering",
        "Aerospace Engineering",
        "Physics"
      ],
      "related_proj": [
        168,
        182,
        60
      ]
    },
    {
      "student_name": "Neha K.",
      "mentor_name": "Udgam Goyal",
      "mentor_title": "MEng CS from MIT, Product Manager at Aurora and prior AI Product Management at Microsoft",
      "mentor_image": "assets/images/mentor_imgs/a3638f2f762e49748ac4a432468a44b6f5a7457b.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Leveraging AI to Analyze Factors Relating to Social Anxiety",
      "project_desc": "With our changing lifestyles and the rise of technology, there has been a steady rise of reported social anxiety cases throughout the years. Exploring how different factors contribute to social anxiety disorder may help people understand significant causes in a cost effective and relatively reliably way. Our research aims to leverage ML to understand what factors play the most critical role in the presence of social anxiety disorder in a person. Additionally, through training different models, we aim to determine the capability of AI in predicting one’s social anxiety. To analyze the factors relating to social anxiety, we clustered our data to find relationships within the background data (e.g: family history, age, gender, symptoms) and one’s social anxiety diagnosis. Our findings suggest that fears and physical conditions may contribute more to social anxiety than a person’s history/background.\n",
      "research_paper": "assets/pdfs/fec47311105c4fafa66632e4f9a04769365fa96f.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://images.onlymyhealth.com/imported/images/2023/March/22_Mar_2023/Signs-of-Social-Anxiety-Disorder-Main.jpg",
      "headline": "Our research aims to leverage ML to understand what factors play the most critical role in the presence of social anxiety disorder in a person.",
      "project_id": "94ad35d1a84b38252fe686e8edb852af014b9ac5",
      "expand": false,
      "tags": [
        "Mental Health"
      ],
      "related_proj": [
        73,
        157,
        59
      ]
    },
    {
      "student_name": "Maryam A.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": " A Logistic Regression Model for Intraoperative Hypotension Prediction",
      "project_desc": " This paper presents a study on using a logistic regression model for the prediction of intraoperative hypotension, a common but critical situation occurring during surgeries where the patient's blood pressure drops significantly. The model was developed using the Python scikit-learn library and was trained on 80% of the available data, while the remaining 20% was used for testing and evaluation. After preprocessing the data, the logistic regression model was trained and then predicted the occurrence of hypotension. The model demonstrated high accuracy, with 97.92% for predictions one minute ahead, but a slight decrease in accuracy was observed as the prediction time increased. When personalized for this type of surgery, the model's performance improved for only a few surgeries, otherwise not highlighting the efficacy of an individualized approach. However, the model faced limitations in predicting hypotensive events beyond a few minutes in advance, necessitating future exploration of more advanced machine learning models. While the logistic regression model's performance was promising, other machine learning models like gradient boosting and random forest algorithms also show high potential. The research underscores the potential of machine learning models in enhancing patient safety and optimizing surgery outcomes through accurate prediction of intraoperative hypotension.This paper presents a study on using a logistic regression model for the prediction of intraoperative hypotension, a common but critical situation occurring during surgeries where the patient's blood pressure drops significantly. The model demonstrated high accuracy, with 97.92% for predictions one minute ahead, but a slight decrease in accuracy was observed as the prediction time increased. When personalized for type of surgery, the model's performance improved for only certain surgeries, not showing a significant improvement overall. While the logistic regression model's performance was promising, other machine learning models like gradient boosting and random forest algorithms also show high potential. ",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://images.squarespace-cdn.com/content/v1/630cfc051dc6e9071a21edea/b405c6f6-cf47-4db0-8f59-4812d3d84fb5/blood-pressure-shingletown.jpg",
      "headline": " This paper presents a study on using a logistic regression model for the prediction of intraoperative hypotension, a common but critical situation occurring during surgeries where the patient's blood pressure drops significantly.",
      "project_id": "2ff75b570ebbe7dd84c3619cff2c5a92955a9956",
      "expand": false,
      "tags": [
        "Heart Disease",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        157,
        31,
        90
      ]
    },
    {
      "student_name": "Yi-Chen (.",
      "mentor_name": "Samar Abu Hegly",
      "mentor_title": "BS in Electrical Engineering & Computer Science and Neuroscience from MiT",
      "mentor_image": "assets/images/mentor_imgs/1f12946a950e1f40195fdfe6f6aef0297455e7ae.png",
      "student_image": "assets/images/student_imgs/2ac397d7439987c4f5a7b4aa95ac32afd427ed5d.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Detecting Retinal Detachment using images with Machine Learning",
      "project_desc": "Retinal detachment occurs when your retina (a light-sensitive layer of tissue in the back of your eye) is pulled away from its normal position at the back of your eye. Symptoms include increased eye floaters, sudden flashes of light in vision, and dark shadows in the middle or on the sides of vision. It can eventually lead to blindness. Therefore, it is important that we detect it early on and treat it. This research is on how AI and machine learning can be used to detect retinal detachment using images. Before building my models, I did data pre-processing on retinal images data from four different hospitals. Then I built several different machine learning models, including Logistic Regression, Ridge Classifier, Decision Tree Classifier, K-nearest neighbor, Random Forest Classifier, SVM, MLP Classifier, Neural Networks, and Transfer Learning. I then assessed their accuracy using certain metrics. After building the models, I calculated the accuracy score, precision score, recall score, F1 score, and built Confusion Matrices for all of the models. Our results showed a very high accuracy in all of them; they were all in the 90% to 99% range. Therefore, I can assume that Retinal Detachment can be diagnosed using Machine Learning methods. In summary, retinal detachment is a serious issue and needs to be detected and treated. Using our machine learning models, I achieved great results.\n",
      "research_paper": "assets/pdfs/c979a0827af1b08fdee47d3bd2ce6898b0dc0867.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://www.nvisioncenters.com/wp-content/uploads/close-up-brown-eye-960x729.jpg",
      "headline": "This research is on how AI and machine learning can be used to detect retinal detachment using images.",
      "project_id": "4665fd467d2e84261599d6cea1d41ec877ab3237",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Eyes",
        "Physiology"
      ],
      "related_proj": [
        115,
        19,
        73
      ]
    },
    {
      "student_name": "Shawn Z.",
      "mentor_name": "Shreyas Ramesh",
      "mentor_title": "MS in Robotics at UPenn",
      "mentor_image": "assets/images/mentor_imgs/90c4b479c03699a2614db354382b4dae71dcd07c.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Arts and Music"
      ],
      "project_title": "Relationship of Factors that Determine Tweet Virality",
      "project_desc": "How can Artificial Intelligence predict the popularity of posts on Twitter (X)? Twitter’s development in mainstream media has been profound, and as the platform increases more overtime, so does the need for predicting virality of tweets on the platform for opinionation. This paper addresses a growing dilemma: As Twitter becomes ever more influential in the modern landscape, predicting what will go viral can help in the discovery and pushing of new ideas and mitigation of risks associated with online networking. We provide an extensive analysis over a dataset consisting of user data and tweet data to compute a virality rating. Using a learning model, we managed to achieve a maximum of 67.32% accuracy on predicting virality based on all user data based on a 1-5 scale ranking.",
      "research_paper": "assets/pdfs/391ae8675c5617b04e884db39fd6338c1a311424.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://img-cdn.inc.com/image/upload/w_600,ar_16:9,c_fill,g_auto,q_auto:best/images/panoramic/twitter-illo_527099_xwxfzh.webp",
      "headline": "This paper addresses a growing dilemma: As Twitter becomes ever more influential in the modern landscape, predicting what will go viral can help in the discovery and pushing of new ideas and mitigation of risks associated with online networking.",
      "project_id": "5f1a3d3e9d9998bc1f28e25ba5e69a2181325671",
      "expand": false,
      "tags": [
        "Social Media",
        "Prediction"
      ],
      "related_proj": [
        102,
        176,
        92
      ]
    },
    {
      "student_name": "Sriram S.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "A Novel ML Approach to Detect Possible Protein Allosteric Sites For Drug Discovery",
      "project_desc": "In recent years, allosteric sites on proteins, located away from the primary active binding site, have gained prominence as promising drug targets because of the significant control they provide over biological pathways. The process of identifying allosteric sites through methods such as x-ray crystallography is often time-consuming, where the protein structure and the structure of potential allosteric-site binding molecules are analyzed to discover the allosteric site. Thus, machine learning models have been applied to expedite the process of allosteric drug discovery by contributing to allosteric site identification. In this study, we investigated different machine learning models and evaluated their performance, and found that the best models were the random forest classifier and a 5-layer neural network. The random forest had an accuracy of 98.89%, a precision score of 98.20%, a recall score of 99.69%, an F1 score of 98.94%, and an AUC of .9991. The neural network had an accuracy of 97.70%, a precision score of 99.15%, a recall score of 96.04%, an F1 score of 97.57%, and an AUC of .9939. The high accuracy of these models along with their efficiency in running could potentially help the pharmaceutical industry identify these allosteric sites which would help with the allosteric drug manufacturing process.",
      "research_paper": "assets/pdfs/e3211479cbbfa3e0ad0e11209051b92320fcd7a4.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://assets.technologynetworks.com/production/dynamic/images/content/353812/screening-strategies-used-in-drug-discovery-353812-960x540.jpg?cb=12554772",
      "headline": "In this study, we investigated different machine learning models and evaluated their performance, and found that the best models were the random forest classifier and a 5-layer neural network.",
      "project_id": "615b2c2126920d26e911cc4e86a6de0e61af3cd5",
      "expand": true,
      "tags": [
        "Drugs",
        "DNA"
      ],
      "related_proj": [
        24,
        108,
        184
      ]
    },
    {
      "student_name": "Orhan H.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/26f40d63c624793bde5380a5f791a6a5fc23d705.png",
      "domains": [
        "Misc"
      ],
      "project_title": "Rare Coin Grading using Convolutional Neural Networks",
      "project_desc": "Rare coins differ greatly in value based on their quality, leading to the development of a standardized 1-70 scale on which coins are graded. The grading process, though, is time and resource-intensive, making it far too expensive for a casual coin collector to get a coin graded. To democratize this process, we developed a computer vision model that grades coins based on an image alone, which makes grading a coin essentially free of cost and provides a dramatic reduction in time. Our dataset consisted of labeled image data from Professional Coin Grading Services, on which we applied affine geometric data augmentations to mitigate possible domain shift on new, “imperfect” images. We experimented with training several different convolutional neural network architectures on this data. Furthermore, we utilized several metrics to effectively evaluate and compare different coin grading models. Our final accuracy was over 50%, and 77% of the predictions were within one grade of the correct grade, affirming that a model sufficiently performant for consumer use is possible. Future work could involve crowdsourcing natural datasets of coin images and making a web or mobile application interface to make our model available to the public.",
      "research_paper": "assets/pdfs/6c5cc7ce666939247e708233f4cdf25f4a73a21e.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://s.yimg.com/ny/api/res/1.2/Los6mzCRSjJasrN3ToweBw--/YXBwaWQ9aGlnaGxhbmRlcjt3PTY0MDtoPTM2MA--/https://media.zenfs.com/en/gobankingrates_644/8568c31625dea6e5aa8f75163bedfd92",
      "headline": "The grading process, though, is time and resource-intensive, making it far too expensive for a casual coin collector to get a coin graded. To democratize this process, we developed a computer vision model that grades coins based on an image alone, which makes grading a coin essentially free of cost and provides a dramatic reduction in time.",
      "project_id": "78f0e01f28258ca04624c000611c99f0837a9147",
      "expand": false,
      "tags": [
        "CNNs",
        "Product Analysis"
      ],
      "related_proj": [
        139,
        42,
        11
      ]
    },
    {
      "student_name": "Sid S.",
      "mentor_name": "Yin Kwong John Lee",
      "mentor_title": "Statistics Lecturer at UChicago, MS in Data Analytics",
      "mentor_image": "assets/images/mentor_imgs/d096d47afb3f61c54536bf220ecc7a2705c6a722.png",
      "student_image": "assets/images/student_imgs/bd5bf99d99cd40b4f86ac11e60471b4fde32b15b.png",
      "domains": [
        "Misc"
      ],
      "project_title": "Relating Coffee Species to Brew Taste and Region with Machine Learning",
      "project_desc": "Buying coffee beans could be a confusing task, due to the large variety, each with their own region and flavor profiles. For many consumers this could lead to overbuying, or making unwanted purchases. If coffee flavor could be correlated to species and even country of origin, coffee purchasing would be much simpler, as consumers could use their own preferences to determine what to buy. This study aimed to find that correlation, using Data from the Coffee Quality Institute, taken from kaggle, and make predictions on the species based on the taste characteristics of the beans. Due to limitations with the data, and keeping in mind that the models tested weren’t tuned to perform their best, the Random Forest model performed best, having one of the highest accuracies with the most favorable set advantages and disadvantages. ",
      "research_paper": "assets/pdfs/c96f488798ea7651f2cd89719712198a20d0672b.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://harweldenmansion.com/wp-content/uploads/2023/09/Coffee-.webp",
      "headline": "If coffee flavor could be correlated to species and even country of origin, coffee purchasing would be much simpler, as consumers could use their own preferences to determine what to buy. This study aimed to find that correlation, using Data from the Coffee Quality Institute, taken from kaggle, and make predictions on the species based on the taste characteristics of the beans.",
      "project_id": "25baa7faf331188083c3a5d51e154e31ce742e9a",
      "expand": false,
      "tags": [
        "Product Analysis",
        "Food and Beverage"
      ],
      "related_proj": [
        138,
        61,
        90
      ]
    },
    {
      "student_name": "Joann C.",
      "mentor_name": "Ivan Felipe Rodriguez",
      "mentor_title": "PhD Candidate at Brown",
      "mentor_image": "assets/images/mentor_imgs/0e30c36e4e6916543bba15053026d44332d58185.png",
      "student_image": "assets/images/student_imgs/c340b0371c12550719b6029b16dca80b30e91dd4.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Neural Radiance Fields (NeRF) for 3D Visualization Rendering Based on 2D Images",
      "project_desc": "We are living in a 3D world, but oftentimes we visualize things in 2D scopes like in pictures and videos of different objects. The world of 3D visualization has been constantly evolving over the past few years, still growing with the introduction of cutting-edge computer vision and deep learning methods. Reconstructing 3D scenes and being able to interact with them in virtual environments used to be something beyond our capabilities but now with the use of Neural Radiance Fields (NeRF) and Instant Neural Graphic Primitives (InstantNGP) we are able to create a world in a virtual space, opening the gateways to applications in object detection and generation, augmented reality, medical imaging, and more. Access to these methods will allow for more immersive development by creators, hands-on educational resources, and more experimentation and research, stimulating future applications and techniques.",
      "research_paper": "assets/pdfs/a9861d7cf5daa36bff27a5edf24a3230abaaeeec.pdf",
      "project_yr": "2022",
      "project_quarter": "Winter",
      "graphic_link": "https://smarttek.solutions/wp-content/uploads/3dmodelling-services-1170x516.jpg",
      "headline": "Reconstructing 3D scenes and being able to interact with them in virtual environments used to be something beyond our capabilities but now with the use of Neural Radiance Fields (NeRF) and Instant Neural Graphic Primitives (InstantNGP) we are able to create a world in a virtual space, opening the gateways to applications in object detection and generation, augmented reality, medical imaging, and more.",
      "project_id": "1a50abdfd8876387fa2c410e2db1a71e4ff8becc",
      "expand": true,
      "tags": [
        "Engineering",
        "Physics"
      ],
      "related_proj": [
        130,
        99,
        27
      ]
    },
    {
      "student_name": "Vihaan B.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/7dcd9ef5296c78dad0d6450c37d41982d45acb45.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Drone Obstacle Detection Using YOLOv5",
      "project_desc": "This project was built with the goal of performing real-time obstacle detection on a drone. In order to achieve this given the lack of pre-existing drone image datasets, I had to collect images, label them, and train a functional model. I collected my dataset using the DJI Mavic Mini Drone and labeled the images on Roboflow, a computer vision data labeling website. I resorted to creating my own dataset focused around 6 different classes: buildings, houses, fences, poles, street signs, and trees. Based on these goals, I picked 3 locations to take pictures: Brandywine Drive, Saratoga High School, and downtown Los Gatos. I flew my drone at a varying range of altitudes from around 5-30 meters off the ground and altered the angle of the gimbal from looking straight down (90°) to horizontal views at around (10°). Throughout the process of labeling, I did not label minute objects which did not directly affect the flight path of the drone. Hence, I did not label many trees/buildings/poles/etc. in the background where I did not find it necessary to do so. Data augmentation has been key to the success I have had throughout this project. Labeling is a lengthy process requiring excessive labor, and a quick way to easily upscale the size of my dataset was through augmentation. I used Roboflow’s threefold (3x) augmentation and ended up settling on 4 types of image-level augmentation: noise, exposure, brightness, and blur. I trained my model using transfer learning on YOLOv5, and was able to achieve proficient results. Albeit mediocre training accuracies, my models performed well on test datasets, correctly creating bounding boxes with high confidences. In the near future, I will look to increase the size of my dataset and balance the label count in order to improve my model and address the skew in my dataset. Additionally, I intend to open-source my current dataset along with my best model on Kaggle and Github to help others dive into the relatively unexplored domain of drone AI.",
      "research_paper": "assets/pdfs/166ccc444e009398fd1ec6a8335ccb0cbc614693.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "github": "https://github.com/vihaanb-v/yolov5",
      "graphic_link": "https://cdn.mos.cms.futurecdn.net/kXUY9hyetVzhZ2scwJP7p3-1200-80.jpg",
      "headline": "This project was built with the goal of performing real-time obstacle detection on a drone.",
      "project_id": "4cbd7912aa2b26a6e01fce8033e84450cc8b7ab5",
      "expand": false,
      "tags": [
        "Robotics",
        "Engineering",
        "Computer Vision"
      ],
      "related_proj": [
        180,
        176,
        30
      ]
    },
    {
      "student_name": "Seohyun P.",
      "mentor_name": "Kasra Koushan",
      "mentor_title": "MSE in CS from Penn",
      "mentor_image": "assets/images/mentor_imgs/bf7b01506cfd765c5e6a874e944f0a83beb5275b.png",
      "student_image": "assets/images/student_imgs/23195f0d484ffc934772a1ca4c5b191d2ad97c8a.png",
      "domains": [
        "Sports and Motion Sciences"
      ],
      "project_title": "Comparison of Machine Learning Models to Best Predict Game Attendance in Major League Baseball",
      "project_desc": "To forecast Major League Baseball game attendance, this study employs six different regression models commonly used for machine learning. The models include Multiple Linear Regression, Decision Tree, Random Forest, Support Vector Regression, Multi-Layer Perceptron Regression, and Gradient Boosting. For the dataset, the 2022 and 2023 games of the Los Angeles Dodgers are used to reflect the recent trend of fan attendance and focus on a single random team so that different tendencies of each team’s fans do not become a confounding variable. Each regression model’s performance is evaluated by Mean Absolute Percentage Error and Root Mean Square Error. The performance evaluation suggests that Random Forest and Gradient Boosting predict attendance with the highest accuracy. Using these prediction models, each baseball team may facilitate staff management, event organization, and marketing before their games.",
      "research_paper": "assets/pdfs/08c7c0f236069ad93e62817e5b62ce51038e5be7.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://cdn.vox-cdn.com/thumbor/yhsudfZl10HwD0hfQVaukGhGE6k=/1400x1050/filters:format(jpeg)/cdn.vox-cdn.com/uploads/chorus_asset/file/13673706/1043223956.jpg.jpg",
      "headline": "To forecast Major League Baseball game attendance, this study employs six different regression models commonly used for machine learning.",
      "project_id": "82e46a61593e6d202f711cefe72392ea5a32be0b",
      "expand": true,
      "tags": [
        "Sports",
        "Baseball",
        "Prediction"
      ],
      "related_proj": [
        30,
        11,
        4
      ]
    },
    {
      "student_name": "Rohan B.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/c600fee6dd2c31917b49d9a13309902713b158fe.png",
      "domains": [
        "Healthcare and Biology",
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Computational Approaches to Enhance Idiopathic Intracranial Hypertension Diagnosis: A Neural Network-Based Framework for Improved Clinical Decision Support",
      "project_desc": "Idiopathic Intracranial Hypertension (IIH) is a condition that needs experts to confirm the diagnosis correctly. It's a long-lasting problem where the pressure inside your head is too high, and if left untreated, it can harm your optic nerve permanently and become tougher to handle. Making a reliable way to diagnose IIH is crucial. This helps to reduce mistakes made by people and makes treating the disorder easier.\nIIH is chronic, meaning it lasts a long time. If it's not treated, it can cause damage to the optic nerve, which is essential for vision. The longer it goes untreated, the harder it becomes to treat. This makes finding an accurate way to diagnose IIH early on very important.\nA computer model, trained on eye images, can tell if optic nerves are healthy or not with 97% accuracy. This is important because it messes up less often than people do, making it a useful tool. The model looks at pictures of the back of the eye (fundus imaging) to make these predictions.\nReducing the number of misdiagnoses is a big deal. With a higher accuracy of diagnosing IIH, we won't need experts to approve as much, and we can be better at planning how to treat it. This can lead to more efficient and effective treatment plans for individuals dealing with IIH.",
      "research_paper": "assets/pdfs/fe3241fdba402d53dedd2a90b00b1e88bcff16dd.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://www.neurosurgeonsofnewjersey.com/wp-content/uploads/neurosurgeonsofnewjersey-what-is-an-idiopathic-intracranial-hypertension.webp",
      "headline": "A computer model, trained on eye images, can tell if optic nerves are healthy or not with 97% accuracy. This is important because it messes up less often than people do, making it a useful tool. The model looks at pictures of the back of the eye (fundus imaging) to make these predictions.",
      "project_id": "754c784f63af9a35b8e6230f4746b04c85e674fa",
      "expand": true,
      "tags": [
        "Neuroscience",
        "Eyes",
        "Computer Vision"
      ],
      "related_proj": [
        19,
        58,
        167
      ]
    },
    {
      "student_name": "Lucas C.",
      "mentor_name": "Ross Greer",
      "mentor_title": "PhD Candidate in Electrical and Computer Engineering, Research at MIT and Berkeley, Engineer at Tesco",
      "mentor_image": "assets/images/mentor_imgs/c995a8ff1784ed78d976edf101087772c99899fa.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Engineering",
        "Arts and Music"
      ],
      "project_title": "Human-Robot Task Handoff: A Probabilistic Modeling Approach Explored through Cooperative Drawing",
      "project_desc": "Recent research in human-robot interaction explores the potential for human-machine collaboration in surgical procedures, dividing tasks into manual and automatable subtasks. This paper investigates the task of handoff detection, crucial for the success of robot-assisted surgery, focusing on the creation of a synthetic dataset that can be used for training and benchmarking models for this task. We present a dataset of parabolas, simulating cooperative drawing between a human and a robot, with variations in drawing rates and added noise. The study demonstrates the applicability of HMMs in determining handoff points, laying the groundwork for future research in human-machine collaborative surgery. The dataset, along with the provided code and raw data, are provided as a resource for future research. Finally, we discuss the limitations of the dataset and suggest directions for future research, emphasizing the need for higher-dimensional and real-world datasets.",
      "research_paper": "assets/pdfs/30d5d8062790126745edd866677719b0196b39a1.pdf",
      "project_yr": "2022",
      "project_quarter": "Fall",
      "github": "https://github.com/lucasc073/GMM1/blob/main/ParabolasWithNoise.ipynb",
      "graphic_link": "https://www.medicaldevice-network.com/wp-content/uploads/sites/23/2020/03/cmr-surgical-versius-clinical-setup-4-arms-00013-2.jpg",
      "headline": "This paper investigates the task of handoff detection, crucial for the success of robot-assisted surgery, focusing on the creation of a synthetic dataset that can be used for training and benchmarking models for this task.",
      "project_id": "fcbf889b2c91cfc127e896309a0c9462b1daff5f",
      "expand": true,
      "tags": [
        "Robotics",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        128,
        96,
        130
      ]
    },
    {
      "student_name": "Anika S.",
      "mentor_name": "Ying Hang Seah",
      "mentor_title": "MS in Management Science and Engineering and BS in Computer Science from Stanford, Startup Co-founder",
      "mentor_image": "assets/images/mentor_imgs/903661425b47b21952595049ff93424d134d2506.png",
      "student_image": "assets/images/student_imgs/3e81a81724cf0073d0800f5d96bcfd5615110f17.png",
      "domains": [
        "Environment"
      ],
      "project_title": "Landfill Net: A Convolutional Neural Network (CNN) Architecture for the Detection of Landfills from Satellite Imagery in the Continental United States",
      "project_desc": "As we continue through the 21st century, the effects of climate change have become more and more apparent, with methane emissions playing a pivotal role as one of the most potent greenhouse gasses. Landfill decomposition is a large contributor to this rise in methane, and certain landfills have even adopted methane capture systems that enable the conversion of this methane into valuable energy resources. To continue to mitigate these emissions, it is imperative to establish a comprehensive United States landfill database. We present a Convolutional Neural Network (CNN) model that uses image data in the United States to distinguish whether an image is a landfill or not. The model had a test accuracy of 97.1%, a precision of 96.3%, a recall of 98.1%, and an F-1 score of 97.2%. The results show that the model holds the potential to uncover illegal landfills within the United States and facilitate an informed expansion of methane collection systems in landfill sites.",
      "research_paper": "assets/pdfs/9b7a9bb8993cb8645f1200f22b54f5e7d19beefc.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "github": "https://github.com/anikaSeshan/LandfillNetResearch.git",
      "graphic_link": "https://www.plasticexpert.co.uk/wp-content/uploads/2022/11/pexels-tom-fisk-5424854.jpg",
      "headline": "We present a Convolutional Neural Network (CNN) model that uses image data in the United States to distinguish whether an image is a landfill or not.",
      "project_id": "cb6efe28ac166c1553671ee2e20ce03f130ee44f",
      "expand": true,
      "tags": [
        "Environmentalism",
        "Computer Vision",
        "Pollution",
        "Sustainability"
      ],
      "related_proj": [
        33,
        27,
        107
      ]
    },
    {
      "student_name": "Dhaniya V.",
      "mentor_name": "Matan Gans",
      "mentor_title": "Computer Science MS from Brown",
      "mentor_image": "assets/images/mentor_imgs/73681a9d5dfa8c731c6630f956bd88bbcea828bc.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Language and Literature"
      ],
      "project_title": "Is That Slang?",
      "project_desc": "Everyday language rapidly changes as the meanings of words are contextualized to fit certain cultural or demographic needs.  There are also words that have double meanings which, depending on the context, can determine whether a sentence has Gen-Z slang in it or not.  For example, ‘The stars lit up the sky.’ and ‘This party is lit!’ both use the word ‘lit’ but only the second sentence is considered Gen-Z slang.  This can be very confusing for people of the older generations to understand as some sentences contain slang that does not make sense.  How can we better understand Gen-Z slang that is being used today?  In order to tackle this challenge, I decided to make a NLP model that processes sentences and determines whether or not they contain Gen-Z slang.  Using Logistic Regression in particular, I was able to create a model that processes inputted sentences as well as went over methods to checking accuracy.",
      "research_paper": "assets/pdfs/762810ca36bb8d4a59ff7bcc866052d79d7edbc5.pdf",
      "project_yr": "2024",
      "project_quarter": "Spring",
      "graphic_link": "https://static.studyusa.com/article/cdn_pLZCrA8qowQjJW7jLyLju4Uh7TypEb3I_sm_2x.jpg?format=webp",
      "headline": "How can we better understand Gen-Z slang that is being used today?  In order to tackle this challenge, I decided to make a NLP model that processes sentences and determines whether or not they contain Gen-Z slang.",
      "project_id": "5a78c5130a64bedf75d1697bf6fc9e48fe030de6",
      "expand": false,
      "tags": [
        "Natural Language Processing",
        "Linguistics"
      ],
      "related_proj": [
        51,
        38,
        162
      ]
    },
    {
      "student_name": "Ronit M.",
      "mentor_name": "Hassan Azmat",
      "mentor_title": "MS in Mechanical Engineering from CMU, startup co-founder",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/eee3de32ac6adbe66678c46b33814cccc8f50506.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Predicting Fake Job Listings  from Real Ones using Machine Learning Models",
      "project_desc": "Nowadays, there are more and more fake job positions appearing online, whether that be companies listing more job positions than they actually need or malicious actors listing out job positions that are meant to take personal information. The main way this problem is going to be solved is through the creation of models that can predict whether these job listings are real or fake. The procedure involves finding the dataset needed to train the model. After the preprocessing, the different NLP models were tested out to see which one would end up giving the best prediction, with the only ones that  ended up  working in the end being a Logistic Regression Model and a BERT Model. The accuracy of the Logistic Regression model ended up being 99.48%,. And after running the data through Random Forest Hyperparameter Tuning, the model was left with a mean absolute error of 3.37% (low values are good!). The accuracy of the BERT model ended up being 99.81%. Overall, the accuracy of the logistic regression model on its own was excellent in separating the fake job listings from the real ones, showing its capability in that regard. The BERT model was even better, with its increase by 0.33% in terms of accuracy. Though, possible potential ways to further improve the accuracy to a greater level would be through  implementing a LSTM model, given how it can be more accurate than Logistic Regression and as accurate as a BERT model.Though, both require their own amount of preprocessing of the data that is separate from the main preprocessing already completed. Alongside that, it would be beneficial to have the models be able to take in input and be easily accessible so that anyone who is concerned about whether or not a job position is real or fake can check through imputing the information into a search box and submitting it, giving the models more data to use to improve their accuracy along with giving the public peace of mind to know that their data isn't going to get stolen. ",
      "research_paper": "assets/pdfs/e41bc1e2b1380906e267fd864dea050181794d8f.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://d3kqdc25i4tl0t.cloudfront.net/articles/content/449_566610_linkedin.hero.jpg",
      "headline": "The main way this problem is going to be solved is through the creation of models that can predict whether these job listings are real or fake.",
      "project_id": "e7bd3704175e954c53553307d6aeb08e983bf8ff",
      "expand": false,
      "tags": [
        "Economics",
        "Fraud"
      ],
      "related_proj": [
        122,
        88,
        89
      ]
    },
    {
      "student_name": "Ryan C.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Deep Learning Approach to Gated Coronary Artery Calcium Scan Segmentations",
      "project_desc": "Coronary artery calcium (CAC) tests are a CT scan of the heart that takes detailed images of arteries to locate deposits of calcium. Excessive calcium in the arteries can lead to reduced blood flow to the heart and cause heart diseases. These are important indicators for heart attacks and diseases and getting CAC scans are useful in identifying high-risk individuals. Agatson scores are the calculations of total calcium in the heart, state of the art programs are currently about 95 percent accurate. The work done in this paper highlights the use of Tversky loss in small segmentations and using an ensemble model to achieve higher accuracy. ",
      "research_paper": "assets/pdfs/ec9d8730cc5cab6336c3c6ff7c7c95be53a4eed8.pdf",
      "project_yr": "2024",
      "project_quarter": "Winter",
      "github": "https://github.com/RyanChew1/Chest-CT-CAC-Scores",
      "graphic_link": "https://cdn.mos.cms.futurecdn.net/8Mg2k3Mh6rtJGz8A5cixUG-1200-80.jpg",
      "headline": "The work done in this paper highlights the use of Tversky loss in small segmentations and using an ensemble model to achieve higher accuracy. ",
      "project_id": "ff9640910e2640cfcaf7dd1b6fab69e3738aa2d8",
      "expand": false,
      "tags": [
        "Heart Disease",
        "Deep Learning"
      ],
      "related_proj": [
        163,
        21,
        93
      ]
    },
    {
      "student_name": "Barbara T.",
      "mentor_name": "Ivan Villa-Renteria",
      "mentor_title": "MS Computer Science at Stanford",
      "mentor_image": "assets/images/mentor_imgs/c3367c49c4a121b7eae2189efc7662edd8de4720.png",
      "student_image": "assets/images/student_imgs/5d375d3fca09d5fee4265b6b92a5031bd2c96d9a.png",
      "domains": [
        "Humanitarian Efforts and Social Justice"
      ],
      "project_title": "Telling Gunshots and Gunshot-like Sounds Appart",
      "project_desc": "Sounds differ from each other. Having a strong ML model that can distinguish between the sounds and correctly classify them, may help with a variety of social problems, such as human and animal well being, (e.g. some sounds cause harm to autistic people and/or animals), safety drills (people with hearing disabilities may not be aware of the alarms), mass shootings, etc. Although reducing a false positive rate (e.g. a rate of other sounds falsely classified as gunshots) would help decrease the number of false alarms, it is more important (for the safety and security reason) to actually reduce the false negative rate (e.g. a rate of actual gunshots not being recognized as such). \nThe research started with searching for and collecting data. The first database found in one of the public github repositories is a collection of plastic bag pop sounds. The second database is a collection of various gunshot recordings. An identical number of data points from each database was used as an input to the machine learning models. Two types of ML models have been developed: classification linear regression model and convolutional neural network (CNN). Furthermore, in order to prevent these models from overfitting by learning from only these two different types of sound, the original dataset has been augmented and a random noise was added to the half of this enlarged database.\nThe first classification model (logistic regression) used with the original database resulted in 0.561 accuracy. When used with the augmented dataset with half of it being slightly distorted, its accuracy level was ranging from 0.557 to 0.610. The CNN model, on the other hand, produced much better results with both the original and augmented data. The accuracy level of the convolutional neural network used with the original data reached 0.972 for 14 epochs and 0.996 for 30 epochs, whereas the accuracy level of the CNN model used with the augmented data was 0.983 for 14 epochs and 0.972 for 30 of them.\nWhile CNN model reported back very high accuracy, it is still not good enough for recognizing actual gunshots among a variety of sounds, which were not used in the project, and for this reason it is not meant to be the sole tool for recognizing an active shooter.",
      "research_paper": "assets/pdfs/b58bfef2303e0e03431fbcf6a6d82f12c78e625e.pdf",
      "project_yr": "2024",
      "project_quarter": "Winter",
      "graphic_link": "https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcQpqY7YzA2qCWhqu_fYdZ_RP5Q5ZGp19A6oPQ&s",
      "headline": "Sounds differ from each other. Having a strong ML model that can distinguish between the sounds and correctly classify them, may help with a variety of social problems, such as human and animal well being, (e.g. some sounds cause harm to autistic people and/or animals), safety drills (people with hearing disabilities may not be aware of the alarms), mass shootings, etc. Although reducing a false positive rate (e.g. a rate of other sounds falsely classified as gunshots) would help decrease the number of false alarms, it is more important (for the safety and security reason) to actually reduce the false negative rate (e.g. a rate of actual gunshots not being recognized as such). ",
      "project_id": "d646ae24ff1f03ce15e6ee6a791dfb00d9e98675",
      "expand": false,
      "tags": [
        "Audio Analysis",
        "Classification"
      ],
      "related_proj": [
        133,
        100,
        0
      ]
    },
    {
      "student_name": "Sambhav A.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/b72acc6352ba6fc4ba0709a3f8a6b995a954d029.png",
      "domains": [
        "Physics",
        "Engineering"
      ],
      "project_title": "Predictive Analysis of Aircraft Engine Lifespan: Leveraging Neural Networks on NASA's C-MAPSS Dataset",
      "project_desc": "This research paper explores the application of neural networks for predictive maintenance of aircraft engines, focusing on the estimation of their Remaining Useful Life (RUL). Using the NASA C-MAPSS dataset, a rich resource in predictive maintenance, the study explores the effectiveness of linear regression and neural network models in forecasting engine RUL. The motivation behind this project was to enhance aircraft safety, extend engine lifespan, and optimize maintenance schedules, thereby reducing operational costs. The initial phase involved linear regression to establish a baseline understanding, followed by the development and iterative enhancement of a neural network model. The results indicated satisfactory performance on the initial dataset, such that it performed with little loss, but revealed challenges in generalizing across different datasets, such that it performed with exceedingly more loss compared to the first dataset, pointing towards the need for more adaptive modeling and feature engineering strategies. \n",
      "research_paper": "assets/pdfs/47b132510cfca8477b6c56163ab708cf809bb3a0.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://media.cnn.com/api/v1/images/stellar/prod/190712104202-derby-civil-aerospace-trent-xwb-product-2.jpg?q=w_4342,h_2443,x_0,y_0,c_fill",
      "headline": "This research paper explores the application of neural networks for predictive maintenance of aircraft engines, focusing on the estimation of their Remaining Useful Life (RUL).",
      "project_id": "5c59dc9365f1ae2733b9011ffe344b825a55ebc6",
      "expand": true,
      "tags": [
        "Engineering",
        "Aerospace Engineering",
        "Physics",
        "Hardware"
      ],
      "related_proj": [
        69,
        173,
        130
      ]
    },
    {
      "student_name": "Mariah D.",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/f18663ea6d71a9d45bdcba31695484f9b9cf39c6.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Predictive Dynamics of Solar Cycles: An Analysis of Sunspot Patterns, Granger Causal Relationships, and Terrestrial and Space Weather Phenomena ",
      "project_desc": "This paper attempts to understand the complex nature of solar cycles, primarily through the lens of sunspot data analysis, to predict future solar activity and explore its causality with Earth and Space phenomena such as global temperatures and CO2 Emissions. Using statistical time series analysis methods, including Seasonal AutoRegressive Integrated Moving Average (SARIMA) and Long Short-Term Memory (LSTM) networks, predictive models were developed based on historical sunspot records dating back to 1818 up to 2019. The aim was to create a forecast mechanism to indicate and predict future solar cycle patterns. Granger causality tests were used to investigate the influences of sunspot activity on other variables of interest, such as CO2 emissions and climate temperature variations. The findings of this analysis contribute to the growing research in astrophysics and environmental science to offer insights that could be beneficial for future research into solar phenomena and its effects on Earth.  \n",
      "research_paper": "assets/pdfs/1b76ede2c1a7123e7795fa7bd2d2fd41d3becc07.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://www2.hao.ucar.edu/sites/default/files/styles/extra_large/public/2022-01/slide17.jpg?itok=UdMYpwDh",
      "headline": "This paper attempts to understand the complex nature of solar cycles, primarily through the lens of sunspot data analysis, to predict future solar activity and explore its causality with Earth and Space phenomena such as global temperatures and CO2 Emissions.",
      "project_id": "73b6110a7dfa250dc5206d04d760431ea59ac72e",
      "expand": true,
      "tags": [
        "Physics",
        "Astronomy",
        "Solar Cycles"
      ],
      "related_proj": [
        110,
        60,
        166
      ]
    },
    {
      "student_name": "Akshaj S.",
      "mentor_name": "Kasra Koushan",
      "mentor_title": "MSE in CS from Penn",
      "mentor_image": "assets/images/mentor_imgs/bf7b01506cfd765c5e6a874e944f0a83beb5275b.png",
      "student_image": "assets/images/student_imgs/aa94527b1ab608258313c92895694970dacff2bd.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Predicting the Severity of Alzheimer’s/Dementia from Magnetic Resonance Imaging Scans Using a Deep Learning Approach",
      "project_desc": "Dementia and Alzheimer’s disease are growing health challenges, as an estimated 10 million new diagnoses happen annually, highlighting the need for more efficient diagnostic processes. The lengthy process involves scheduling appointments, pre-screening assessments, and arranging screening appointments, consuming valuable time for healthcare professionals. A streamlined process leveraging computer-based screening could optimize efficiency. In fact, artificial intelligence-based systems also have potential to decrease overall healthcare costs by upwards of 10%, or over $360 billion annually (The Financial Cost of AI in Healthcare, 2023). Deep learning models, with their precision and technical capabilities, offer a promising solution. By automating screening tasks using deep learning, they can reduce administrative burdens, enhance accuracy, and expedite diagnoses, ultimately improving patient care and outcomes. The overall approach used in this research was utilizing a deep learning model, specifically a convolutional neural network. A convolutional neural network is a type of neural network architecture that specializes in identifying and making sense of patterns in image data. For this research, this type of neural network was mostly used for classifying images of dementia MRI scans based on their severities. It used certain classifying methods such as multi-class classification and a high volume of data (roughly 6400 images) to accurately distinguish between different levels of dementia severity. The model's performance metrics are robust, achieving an accuracy of 0.989, a precision of 0.99, a recall of 0.989, and an f1-score of 0.99 after 20 training epochs. These results suggest a high potential for CNN-based models to be integrated into clinical workflows, providing rapid and reliable assessments of dementia severity.",
      "research_paper": "assets/pdfs/a2ababb27b1af99f2d673eb09bd96dd3da2f0a46.pdf",
      "project_yr": "2024",
      "project_quarter": "Winter",
      "graphic_link": "https://genengnews.com/wp-content/uploads/2018/08/Dec22_2016_AJCity_Alzheimer1901775812-1.jpg",
      "headline": "The overall approach used in this research was utilizing a deep learning model, specifically a convolutional neural network. A convolutional neural network is a type of neural network architecture that specializes in identifying and making sense of patterns in image data. For this research, this type of neural network was mostly used for classifying images of dementia MRI scans based on their severities.",
      "project_id": "405c4ba36519ba48408c05c20fc2df7589141e72",
      "expand": true,
      "tags": [
        "Alzheimer's",
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "related_proj": [
        26,
        108,
        73
      ]
    },
    {
      "student_name": "Satvi M.",
      "mentor_name": "James Thomson",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/270aaacb4169c3c3c0828aacf6754b1112d82373.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Predicting Dementia using AI",
      "project_desc": "In this project, I researched and used AI models that predicted if someone is demented or non-demented. The models made its predictions based on several factors such as age, gender, dominant hand, brain size, education, and more. ",
      "research_paper": "assets/pdfs/393259cd288ca873082717e16b071e2cbf7857af.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://d2jx2rerrg6sh3.cloudfront.net/images/Article_Images/ImageForArticle_22616_16540062747895311.jpg",
      "headline": "This project involved researching and utilizing AI models to predict whether an individual is demented or non-demented based on factors like age, gender, dominant hand, brain size, and education.",
      "project_id": "5a7a1c7d976d73a974d8eeb12f79370cd1567e98",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "related_proj": [
        171,
        12,
        50
      ]
    },
    {
      "student_name": "Akshar S.",
      "mentor_name": "John Basbagill",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/d509fcb9e06f76b65b668eb8a9e02a0b3cb005b5.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Predicting Shoe Prices Using Machine Learning Algorithms",
      "project_desc": "This paper explores the application of machine learning algorithms in predicting shoe prices and the correlations between a shoes’ material, color, brand, type, gender, and size and its price. Using a dataset of 5000 shoe entries from Kaggle, the data was preprocessed and turned into numeric values for modeling. The models used include Linear Regression, Decision Tree, Random Forest, Gradient Boosting, and Support Vector Regression. The results show that the Support Vector Regression model had the lowest Mean Squared Error, which means that it can accurately predict shoe prices and find the correlation between a shoes’ make and price.",
      "research_paper": "assets/pdfs/c08c9f21e49228fe23e17aebf3f68e2e4e00f3a4.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://cdn3.vectorstock.com/i/1000x1000/85/27/sneakers-with-blank-price-tag-footwear-shopping-vector-23248527.jpg",
      "headline": "This paper investigates the use of machine learning algorithms, particularly Support Vector Regression, to accurately predict shoe prices based on factors like material, color, brand, type, gender, and size, using a dataset of 5000 entries.",
      "project_id": "24f1dc2e3c24722e6b9fe246cb619983f97a316e",
      "expand": false,
      "tags": [
        "Economics",
        "Pricing"
      ],
      "related_proj": [
        46,
        116,
        122
      ]
    },
    {
      "student_name": "Aditya J.",
      "mentor_name": "Alaisha Alexander",
      "mentor_title": "MS from Stanford, MIT Alum",
      "mentor_image": "assets/images/mentor_imgs/4c906595003fef5a6395866bcb9b4ddf79ff95be.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Melanoma detection through AI models",
      "project_desc": "Skin cancer is one of the most prominent forms of cancer, and early detection can significantly improve the recovery process for those affected. Through this project, I aimed to understand the ability to detect cancer from images of skin lesions using artificial intelligence (AI) models. I attempted to build software which would allow users to predict the possibility of cancerous lesions by processing images through these models. The approach involved the use of different AI algorithms which were trained on a dataset of thousands of individual images, and identifying which models yielded the best results. Just from graphing the data, it was shown that the malignant images were darker on average, with lower average color values for each pixel. The test accuracy of most models ranged between 80 - 90%, proving that the AI models were able to correctly predict malignant v/s non-malignant 80 - 90% of the time. Through my research, I discovered that AI models have the ability to correctly predict melanoma by processing images using a variety of algorithms through which the data is trained. Additionally, I learned of the numerous physical attributes of an image which could help people predict the malignancy of skin lesions before using an AI model for prediction.\n",
      "research_paper": "assets/pdfs/1affd51b6336015187b69955e97e157c1a7cdc08.pdf",
      "project_yr": "2024",
      "project_quarter": "Spring",
      "graphic_link": "https://www.mdanderson.org/images/publications/cancerwise/S/IM_Cancerwise_SkinCheck.jpg",
      "headline": "This project explores the use of AI models to detect skin cancer from images of skin lesions, achieving 80-90% accuracy in predicting malignancy and revealing key physical attributes that could aid early detection.",
      "project_id": "211787ec35ab454019468b40679ef4ec55ab7de1",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Skin Cancer"
      ],
      "related_proj": [
        124,
        6,
        93
      ]
    },
    {
      "student_name": "Aaryan W.",
      "mentor_name": "Anjali Singh",
      "mentor_title": "MIT CS Alum, Research at MIT Media Lab",
      "mentor_image": "assets/images/mentor_imgs/ea6135d921fb4e648f80edb944175526a4120dd3.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "The Impact of Computer Vision on Assisting Visually Impaired People",
      "project_desc": "In today’s world, artificial intelligence has been a significant new innovation. One of the prime aspects of artificial intelligence is its ability to perform computer vision, which is using machine learning to allow a computer to detect things, via camera or sensors. Computer vision has seen significant development in the past years which can be used to help people who struggle with vision. This research focuses on developing computer vision algorithms to output positioning data for people to see. It includes several steps in the coding process, such as creating an object detection model, a positioning algorithm, and finally manipulating videos to project an output. The goal of this research will be to create a working program that can take a sample video and output correct positioning instructions for a user. This work will hopefully transform the way the visually impaired can navigate the world.",
      "research_paper": "assets/pdfs/b268096084009995d5e3ad4e13cee750c7fbc0cf.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://img.freepik.com/free-vector/hand-drawn-people-with-disabilities-illustration_23-2149677954.jpg",
      "headline": "This research aims to develop computer vision algorithms that can provide accurate positioning data from video input, with the goal of creating a program to assist visually impaired individuals in navigating their environment.",
      "project_id": "943e6532b333974920b1d1e107a54561a1ccb859",
      "expand": false,
      "tags": [
        "Biotechnology",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        143,
        97,
        167
      ]
    },
    {
      "student_name": "Anirudh C.",
      "mentor_name": "Matthew Radzihovsky",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/b399999d6df11ae4fa206c4b19330c1439744366.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Machine Learning Models for Cardiovascular Disease: A Holistic Review on Early Diagnosis Performance",
      "project_desc": "Cardiovascular disease (CVD) poses a significant threat to global health, responsible for approximately one-third of all deaths worldwide. Early diagnosis plays a crucial role in preventing severe complications, yet traditional methods often require expensive tests and equipment. Machine learning offers an innovative and more cost-efficient alternative, through pattern recognition of more lightweight data to accurately predict CVD presence and severity. In this paper, we look past traditional binary approaches to CVD diagnosis and look towards quantitative ordinal labels, where a higher number represents more severity. Using the UC Irvine Heart Disease dataset, we test six lightweight models — Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbors (KNN), Multilayer Perceptron (MLP), Support Vector Machines (SVC), and Decision Trees (DT) — alongside a custom neural network model to evaluate their performance on this multi-class classification task. Each model's accuracy, precision, recall, and F1 score are analyzed to assess robustness in clinical settings. The RF, KNN, and custom models achieved testing accuracies above 86%, with the KNN leading in accuracy, precision, and recall, going beyond the baseline LR model’s accuracy of 52%. Our findings demonstrate that lightweight models are highly effective for CVD prediction, offering potential for future deployment in AI-assisted tasks in hospitals. However, these models lack explainability behind diagnosis, which is essential for real-world deployment. Future directions include exploring wearable technology for continuous CVD monitoring and the inclusion of worldwide data to account for diverse risk factors.",
      "research_paper": "assets/pdfs/92e108533f288592a66439c20ecc2656f0e88ef5.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "github": "https://github.com/anirudhc5/cvd-machine-learning-notebook",
      "graphic_link": "https://media.licdn.com/dms/image/D4D12AQGTaolbDut09g/article-cover_image-shrink_720_1280/0/1706783266740?e=2147483647&v=beta&t=Ja3j5IKxMlFszN98O1bb3ZPAKND2wjy-V6kKVjrUrJ4",
      "headline": "This paper evaluates the effectiveness of lightweight machine learning models for predicting cardiovascular disease severity, finding high accuracy and potential for practical use, while emphasizing the need for better explainability and integration with wearable technology.",
      "project_id": "43d31825d4f316dea22f68dae92bb5bbf5f17986",
      "expand": true,
      "tags": [
        "Heart Disease",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        18,
        64,
        68
      ]
    },
    {
      "student_name": "Shreyes B.",
      "mentor_name": "Varsha Sandadi",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Engineering",
        "Specific Technologies"
      ],
      "project_title": "A Late Fusion Approach with Multimodal Image-Text Data",
      "project_desc": "As the number of users on social media platforms continues to rise, an increasing number of people are expressing their emotions and opinions. Before the popularity of multimodal machine learning, previous works included identifying texts and images separately to determine sentiment. However, these studies often overlooked features that other modalities could capture that are crucial to emotion. Recent advances in multimodal machine learning are now making essential and precise judgments in image recognition [1]. In this project, we explore a specific type of fusion called late fusion, demonstrating its ability to combine different modalities. In particular, we highlight two late fusion methods that combine the predictions of the image and text models. Additionally, we investigate a contrastive learning approach, which will help improve the visual embeddings, making them more discriminative in feature space. We compare the strengths and limitations of late fusion against other fusion approaches. Our findings are based on experiments conducted with the publicly available MVSA-Single dataset.\n",
      "research_paper": "assets/pdfs/fb8c976196bebaf887b6af9beac9029de2b351cd.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://sf16-va.tiktokcdn.com/obj/eden-va2/nb-shivsn-ryhs/ljhwZthlaukjlkulzlp/lark-topics/ai-glossary/multimodal-in-machine-learning.webp",
      "headline": "This project explores late fusion methods in multimodal machine learning to improve emotion detection by combining image and text predictions, and evaluates their effectiveness compared to other fusion approaches using the MVSA-Single dataset.",
      "project_id": "45611bdae6df0734251d49fe11aba6314fb664c2",
      "expand": false,
      "tags": [
        "Natural Language Processing",
        "CNNs",
        "Social Media"
      ],
      "related_proj": [
        76,
        132,
        27
      ]
    },
    {
      "student_name": "Sameeksha V.",
      "mentor_name": "Jose Reyes",
      "mentor_title": "UChicago Computer Science Alum",
      "mentor_image": "assets/images/mentor_imgs/c783edc1dcee802fda48afc76dfa7205db18e1df.png",
      "student_image": "assets/images/student_imgs/cf5c62078126e3261b1aa7df89e67cf935453ed5.png",
      "domains": [
        "Education",
        "Engineering"
      ],
      "project_title": "Optimizing Student Success Predictions using Artificial Intelligence",
      "project_desc": "The education field and literacy rates remain popular subjects of discussion within both the research community and the AI field, as people continually seek ways to promote learning and enhance literacy. This work aims to examine the extent of how predictions of students’ success in school can be improved by using AI models and neural networks with respect to student performance, and consequently study more effective resource allocation among students. We concentrate on exploiting artificial neural networks to predict student results; using information about academic history, socio-economic status, attendance rates and engagement levels etc., which inform the allocation of resources like educational materials, teachers, technology devices, & mentors accordingly. We tested different models, including Linear Regression, Logistic Regression and Decision Decision Trees, and neural networks with dense layers. The study finds that using advanced predictive models can greatly increase the types of educational support we offer, reduce disparities in student performance, and address issues revolving around how and where resource allocations may be most equitable. Therefore, we propose a strand of research to mitigate educational resource imbalances and better support low-attainment students with targeted interventions informed by predictive analytics.",
      "research_paper": "assets/pdfs/9ad0ef9b566abf643eb3b1d77cb8ab77c13a8d23.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://f.hubspotusercontent10.net/hubfs/5155768/student-success-journey-blog.jpg",
      "headline": "This work investigates using AI and neural networks to enhance predictions of student success and optimize resource allocation in education, finding that advanced models can significantly improve support for students and address resource imbalances.",
      "project_id": "6800a73e4b57a0964429d3508941ee2c9263cd44",
      "expand": false,
      "tags": [
        "Education"
      ],
      "related_proj": [
        150,
        180,
        158
      ]
    },
    {
      "student_name": "Advik V.",
      "mentor_name": "Ronil Synghal",
      "mentor_title": "UPenn MS Robotics, Data Scientist @ SCPMG",
      "mentor_image": "assets/images/mentor_imgs/b4d4da066f156a8ff549472bd1d2673cf8cde8e1.png",
      "student_image": "assets/images/student_imgs/c455ccfdf49267abea37e9ac5af4a4c67e983118.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Predictive Modeling of Diabetes Using Python Neural Networks",
      "project_desc": "This work uses machine learning approaches to predict the occurrence of diabetes in individuals. Specifically, a neural network model is developed by utilizing the sklearn Python package. Since diabetes is a common chronic metabolic illness, preventing complications and managing the condition effectively depend on early detection. The goal of the research is to accurately classify people as either healthy or diabetic based on pertinent medical characteristics. The model's overall accuracy, balanced performance metrics across the healthy and diabetic classes, and precision, recall, and F1-score are used to evaluate its effectiveness. The outcomes show a respectable performance, suggesting a good potential for diabetes occurrence prediction. The findings emphasize how crucial it is to keep improving the model in order to increase its predictive power and, eventually, help those at risk of diabetes receive better healthcare.",
      "research_paper": "assets/pdfs/734fecfc5c2fccd5a75d209963d24619106f4ac7.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "github": "https://github.com/AdvikVatsyayan/DiabetesPrediction",
      "graphic_link": "https://cdn.vectorstock.com/i/500p/37/76/diabetes-patient-treatment-and-lifestyle-concept-vector-30713776.jpg",
      "headline": "This research develops a neural network model using machine learning to predict diabetes occurrence, achieving strong accuracy and performance metrics, and highlights the need for ongoing model improvements to enhance early detection and healthcare for at-risk individuals.",
      "project_id": "97131fb426c66cb2d7b8ad16b9352f880eb5f976",
      "expand": false,
      "tags": [
        "Diabetes",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        143,
        155,
        21
      ]
    },
    {
      "student_name": "Shishir B.",
      "mentor_name": "Hassan Azmat",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/60375ff67a253b4adfa38f20009bc5630008d843.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Understanding Object Detection’s Role in Decision-Making for Self-Driving Vehicles",
      "project_desc": "Our research aimed to understand computer vision in self-driving vehicles through object detection. We evaluated different AI models to determine the best performer for this task. Self-driving cars, led by companies like Tesla, rely heavily on advanced software, and understanding these systems is crucial before their widespread adoption. We applied machine learning models to detect vehicles in highway surveillance footage, comparing their precision and recall. Among the models tested, VGG16 outperformed others, achieving a precision of 0.7209 and a recall of 0.7298. Object detection in self-driving cars identifies vehicles and other road objects, ensuring safe navigation. Achieving high precision and recall requires large datasets with tens of thousands of images and thorough hyperparameter tuning. Choosing the right model is critical; in our case, VGG16 proved the most effective. These findings underscore the importance of robust model selection and optimization in advanced self-driving technology.",
      "research_paper": "assets/pdfs/3843f4071a210918229f409707486fdc2cb43ac2.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://www.datasciencecentral.com/wp-content/uploads/2021/10/9456892854.jpeg",
      "headline": "This research focuses on computer vision for self-driving vehicles by evaluating various AI models for object detection, finding that VGG16 outperformed others with precision of 0.7209 and recall of 0.7298, highlighting the importance of model selection and optimization in ensuring safe navigation in autonomous driving systems.",
      "project_id": "18352981768f1d001bae3cef68e973a51aa9d1eb",
      "expand": true,
      "tags": [
        "Object Detection Models",
        "Autonomous Vehicles",
        "Engineering"
      ],
      "related_proj": [
        167,
        182,
        141
      ]
    },
    {
      "student_name": "Ian M.",
      "mentor_name": "Alaisha Alexander",
      "mentor_title": "MS from Stanford, MIT Alum",
      "mentor_image": "assets/images/mentor_imgs/4c906595003fef5a6395866bcb9b4ddf79ff95be.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Language and Literature",
        "Generative AI"
      ],
      "project_title": "From Text to Visuals: Leveraging AI for Immersive Storytelling",
      "project_desc": "This research explores the potential of artificial intelligence(AI) to create seamless and customizable visual representations of literary texts, specifically focusing on Edgar Allan Poe’s works. The aim is to bridge the gap between traditional literature and modern digital media by enhancing readers’ visualization of narratives through AI-generated imagery. This is particularly important in the digital age, where engagement with text is evolving due to shorter attention spans and the rise of multimedia content. The methodology involved data preprocessing, including text normalization, tokenization, and sentiment analysis, followed by genre classification using natural language processing(NLP) techniques and a Random Forest classifier. Advanced NLP and image generation models, GPT-3.5-Trubo and DALL-E were then used to create detailed and consistent visual representations of the text. The result demonstrated the models’ ability to generate images that aligned with the narrative and tone of the text, though challenges with consistency and handling complex narrative elements were identified. The study concludes that with further refinement, AI has the potential to transform the reading experience by offering new opportunities for engagement , accessibility, and innovation in the literary space. Future research will focus on experimenting with more advanced models, expanding the dataset, and exploring techniques for maintaining visual consistency across generated images.",
      "research_paper": "assets/pdfs/9e2711a1a5a0038d36f4c20de30b86f7b1757b63.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://images.nightcafe.studio/jobs/RThitBVmEtSoq1Cgh6Zv/RThitBVmEtSoq1Cgh6Zv--1--7bbqo.jpg?tr=w-1600,c-at_max",
      "headline": "This research investigates the use of AI to generate visual representations of Edgar Allan Poe's literary works, aiming to enhance reader engagement by bridging traditional literature with modern digital media through AI-generated imagery, while identifying challenges related to visual consistency and narrative complexity.",
      "project_id": "fb65232bfef7b178fc17bfc88fd18a55b1dd2687",
      "expand": false,
      "tags": [
        "Natural Language Processing",
        "Generative AI",
        "GPT-3"
      ],
      "related_proj": [
        84,
        51,
        118
      ]
    },
    {
      "student_name": "Katie Y.",
      "mentor_name": "Sriram Hathwar",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/699db61c6e75da727bf9abb8050aec8d85da1982.png",
      "domains": [
        "Healthcare and Biology",
        "Engineering"
      ],
      "project_title": "Using Machine Learning to Detect Parkinson’s Disease Through Drawing Data",
      "project_desc": "Parkinson’s is a neurodegenerative disease that causes uncontrolled movements such as tremors or stiffness. Currently, there is no specific clinical test that can test for Parkinsons, therefore many cases go undiagnosed. Patients that do get diagnosed  often must go through a lengthy and expensive process of a series of examinations and consultations with specialists. In this paper, we present a few machine learning models, including logistic regression, K-Nearest Neighbors, Decision Trees, Convolutional Neural Networks, and Transfer Learning, as a potential way to detect Parkinson’s through patients’ drawings of spirals and waves. We decided to use patients’ drawings, as this data is much more accessible and inexpensive to obtain in comparison to imaging or lab results. Coupling this data with machine learning models could therefore serve to increase Parkinson’s diagnoses, particularly in developing countries, and ultimately help patients receive better treatment options. Our results show that the logistic regression classifier performed the best, with  an accuracy of 91%, precision score of 92%, and recall score of 89%. These results show that machine learning and spiral and wave drawings can be used as a viable tool for furthering Parkinson’s diagnosis.",
      "research_paper": "assets/pdfs/e53404ed506183ce3fac9fabb0115de279c4252d.pdf",
      "project_yr": "2024",
      "project_quarter": "Spring",
      "graphic_link": "https://static.vecteezy.com/system/resources/previews/002/613/988/non_2x/parkinson-disease-cartoon-vector.jpg",
      "headline": "This research explores the use of machine learning models, particularly logistic regression, on patients' spiral and wave drawings as an accessible and cost-effective method to improve Parkinson's disease diagnosis, achieving an accuracy of 91%, precision of 92%, and recall of 89%.",
      "project_id": "90afdea6e9eab0f0fe023a7f91e3359d76f205ec",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "related_proj": [
        12,
        171,
        133
      ]
    },
    {
      "student_name": "Trinity M.",
      "mentor_name": "Kevin Phan",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/b2d82c5f8baf1910a1bafab3d539459785fec76f.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Utilizing AI for Alzheimer's Diagnosis",
      "project_desc": "Alzheimer’s is a serious disease and its early diagnosis is essential to properly understand the disease, access information, and provide sufficient care and support. This study hopes to assess the potential of AI to help early diagnoses. When properly used, AI could potentially diagnose Alzheimer’s with more efficiency and accuracy. Many classifier models were used: the MLP classifier had the least accuracy of 32%, the K Nearest Neighbors classifier had an accuracy of 60%, the Decision Tree Model had an accuracy of 72%, the Logistic Regression model had an accuracy of 83%, and the Random Forest classifier did the best with a 91.5% accuracy. There are many factors that the AI algorithm takes into consideration when making the diagnosis, and each factor is weighed differently. Factors like hand dominance had a smaller weight in the diagnosis than the mini mental state evaluation, and all of the analysis is covered in the Logistic Regression Feature Importance to shed light on the factor importance.",
      "research_paper": "assets/pdfs/4f00a47e4381cbf7421e0920b79c4e6904b64f5f.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "project_webpage": "https://colab.research.google.com/drive/1D2Lsohhu6Gejhprll6JtfYIOalHsT_RW#scrollTo=I6kNiV9SYOzy",
      "graphic_link": "https://cdn.vectorstock.com/i/500p/93/97/alzheimer-disease-illness-neurology-therapy-vector-32269397.jpg",
      "headline": "This research assesses the potential of AI to improve early Alzheimer’s diagnosis by evaluating various classifier models, with results showing that while MLP classifier achieved only 32% accuracy, the Random Forest classifier excelled at 91.5%, highlighting the importance of factors like the mini mental state evaluation over others, as revealed in the Logistic Regression Feature Importance analysis.",
      "project_id": "94158a8b9817ca7d811bc605d39e0026cfbcfd6c",
      "expand": false,
      "tags": [
        "Neuroscience",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        155,
        119,
        29
      ]
    },
    {
      "student_name": "Isabella F.",
      "mentor_name": "Simeon Sayer",
      "mentor_title": "Staff Researcher @ Faculty of Arts and Sciences at Harvad, Head Teaching Fellow @ Harvard CS, Harvard CS Alum",
      "mentor_image": "assets/images/mentor_imgs/290a3c8365be2682eacfad1771b06ce529da5d91.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Cybersecurity",
        "PoliSci and Law"
      ],
      "project_title": "Applying Machine Learning to Historical Cyber Attack Data to Predict and Prevent Future Attacks",
      "project_desc": "Geopolitical tensions are high, with instability and uncertainty in multiple conflict zones which ripple around the world to countries that support the rule of law, those that do not, and those that are otherwise affected by the sprawling crises. While these conflicts are catastrophic, they are also convenient cover for state-sponsored actors to launch cyber attacks to damage critical infrastructure, undermine democratic institutions and values, and gather sensitive information to prepare for future attacks. State-sponsored actors have become highly sophisticated, with capabilities that leverage the power of technology to prey on governments and companies. Yet despite this sophistication, their attack patterns often have similar digital fingerprints. This reality creates the possibility that past attacks may provide clues for future ones. This paper, which is based on a machine learning algorithm, finds that historical data alone can be sufficient to predict the nature of future attacks with reasonably reliable accuracy. In particular, the severity of the previous attacks gave reliable insight into the severity of future attacks.",
      "research_paper": "assets/pdfs/349efc75a196757a1be0577c4acdae32d78f2c06.pdf",
      "project_yr": "2023",
      "project_quarter": "Fall",
      "graphic_link": "https://swisscognitive.ch/wp-content/uploads/2021/11/ML_in_Cybersecurity_SwissCognitive_World_Leading_AI_Network_3.jpg",
      "headline": "This research demonstrates that machine learning algorithms can leverage historical data to predict the nature and severity of future cyber attacks with reasonable accuracy, revealing that despite the sophistication of state-sponsored actors, their attack patterns often exhibit digital fingerprints that offer valuable clues for anticipating future threats amidst global geopolitical tensions.",
      "project_id": "04600800bdd26d05dd5543710170b3527ef892cd",
      "expand": true,
      "tags": [
        "Government Systems",
        "Political Science",
        "Engineering"
      ],
      "related_proj": [
        72,
        100,
        78
      ]
    },
    {
      "student_name": "Leonardo V.",
      "mentor_name": "Pablo Bonilla",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/838ab366674f2429062c955080c869f9def1f8c5.png",
      "domains": [
        "Physics",
        "Engineering"
      ],
      "project_title": "Fast and Accurate Gamma-Ray/Hadronic Particle Shower Classification Using Machine Learning",
      "project_desc": "Correct and efficient prediction of atmospheric gamma-ray particle showers is important because the study of these initiating particles can lead us to discoveries about the parts of the universe from which they originated. Previous research has been done with datasets similar to the one we chose, all of which go deep into the optimization of machine learning methods to solve this classification problem. However, we have yet to see researchers try to simplify their model in order to maintain high model performance while minimizing the number of features used. Because of this, we used dimensionality reduction methods like Pearson Correlation and Principal Component Analysis on sklearn classification models like Random Forests and Support Vector Classifiers to try and solve this issue. It was found that with a Pearson Correlation Coefficient cut-off of 0.01, model accuracy was increased from 87.51% to 87.83% while reducing the number of features used by two, meaning our goal was accomplished. In addition, a stacking ensemble yielded an accuracy of 88.04% with fast predicting speed.",
      "research_paper": "assets/pdfs/53f5af7fe79b1e220c5130bc291d98a7e35b26f3.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "github": "https://github.com/LeonardoValli/Particle-Shower-Classification",
      "graphic_link": "https://www.quantamagazine.org/wp-content/uploads/2015/05/CosmicRays_996x581_v2.jpg",
      "headline": "This research applies dimensionality reduction techniques, such as Pearson Correlation and Principal Component Analysis, to machine learning models like Random Forests and Support Vector Classifiers to simplify the prediction of atmospheric gamma-ray particle showers, achieving a modest accuracy increase while reducing the number of features used.",
      "project_id": "b956a677f572edc5e215d0e272858a8ff3fec7b2",
      "expand": true,
      "tags": [
        "Astronomy",
        "Gamma Rays"
      ],
      "related_proj": [
        30,
        163,
        115
      ]
    },
    {
      "student_name": "Sara C.",
      "mentor_name": "Emily Broadhurst",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/556e8c3e002d3b1c14b302549fd5d5b042f97229.png",
      "domains": [
        "Healthcare and Biology",
        "Engineering"
      ],
      "project_title": "Predicting the Presence of Autism Spectrum Disorder Based on Eye-Tracking Scan Path Images",
      "project_desc": "Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects communication, social interaction, and behavior. With over 5.4 million individuals in the United States diagnosed with ASD, early detection is critical for improving outcomes through timely intervention. Eye-tracking technology, which monitors visual attention patterns, has emerged as a potential tool for early diagnosis. Our research aimed to explore whether eye-tracking scan path images could be used to predict the presence of ASD by applying machine learning techniques. We collected eye-tracking data from individuals with ASD and non-ASD controls, processed the images to enhance clarity, and labeled them accordingly (ASD = 1, non-ASD = 0). Several machine learning models were tested, with logistic regression being the most effective, achieving an accuracy of 74.6% using a regularization parameter (C) of 0.1. This indicates that while eye-tracking can aid in ASD classification, there is still room for improvement in model performance. Our findings suggest that eye-tracking combined with machine learning has the potential to assist in early ASD detection. This non-invasive method may also be useful in diagnosing other neurological conditions, such as Alzheimer’s, offering healthcare professionals a new approach for early intervention.\n",
      "research_paper": "assets/pdfs/6c75de2a0808c3233313b55492a2bdf888f2d0bb.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "github": "https://github.com/sarcha734/Autism-Spectrum-Disorder-AI-Classifier/tree/main",
      "graphic_link": "https://neurosciencenews.com/files/2024/08/ai-asd-neuroscience.jpg",
      "headline": "This research explores using eye-tracking data and machine learning to predict Autism Spectrum Disorder (ASD), achieving 74.6% accuracy with logistic regression, highlighting eye-tracking's potential for early, non-invasive ASD diagnosis and its applicability to other neurological conditions.",
      "project_id": "1d602d4f705699f52d148e85ca410e89077fac49",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Engineering",
        "Biology"
      ],
      "related_proj": [
        137,
        173,
        8
      ]
    },
    {
      "student_name": "Akilan P.",
      "mentor_name": "Erick Ruiz",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Physics"
      ],
      "project_title": "Using Deep Learning to Predict the Half-Lives of Isotopes Given Proton and Neutron Count",
      "project_desc": "In this study, we attempt to predict the half-lives of superheavy isotopes through regression models. Superheavy elements usually have very short half-lives, but a few isotopes are predicted to have long enough half-lives for potential applications in nuclear and material science. To calculate these half-lives, we initially used a deep learning regression model with Keras library’s dense layers that took proton count and neutron count as inputs and $\\log_{10}$ of the half-life as the output. When it was tested on various isotopes of the periodic table, there was an MSE (mean squared error) of 111. Afterwards, more input variables related to proton and neutron count were added to increase complexity, leading to an MSE of 97. This suggests that the model on average falls many orders of magnitude off from the actual half-life, which would make accurate prediction of half-lives difficult. It is possible that more complexity, through different kinds of layers, and different inputs may be needed to accurately predict the half-lives of nuclides (another word for isotopes). ",
      "research_paper": "assets/pdfs/ced11a81d6ddbe6086963c7d1a15a567f5333727.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://www.physics.com.sg/images/atomic-vs-nuclear-physics-what-is-the-key-difference.jpg",
      "headline": "This study explores using deep learning regression models to predict superheavy isotopes' half-lives based on proton and neutron counts, finding that additional complexity in input variables and model architecture may be necessary to improve prediction accuracy.",
      "project_id": "80699a13682827c3ad79bf940973d4cbdd6496dd",
      "expand": true,
      "tags": [
        "Astronomy",
        "Physics"
      ],
      "related_proj": [
        91,
        166,
        151
      ]
    },
    {
      "student_name": "Min W.",
      "mentor_name": "John Basbagill",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/aaf28865a6b130217940c758ab5f3adde531e318.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Predicting Dementia Risk with Machine Learning",
      "project_desc": "The rising prevalence of dementia poses a significant global health challenge, particularly as populations age. Given the increasing burden of dementia for individuals, families and healthcare systems, there is an urgent need for innovative research to improve prevention, diagnosis, and treatment. Understanding the risk factors and underlying mechanisms of dementia is critical for developing effective strategies to combat this debilitating disease. Early detection is crucial for managing symptoms and potentially slowing progression. By understanding how medical factors influence dementia risk, doctors may be able to identify individuals who could benefit from early intervention and lifestyle changes to help prevent or delay the onset of dementia. Previous research has explored how factors such as high blood pressure, diabetes, and even personality traits can be linked to an increased risk of dementia [3]. However, these studies often rely on self-reported data or focus on a single factor, making it difficult to untangle the complex causes of dementia. The research took a more comprehensive approach by using a dementia patient characteristics dataset with several medical factors. The approach built machine learning models to predict the relationship between patient health parameters such as heart rate, presence of diabetes, and weight and the likelihood of dementia. Logistic regression and random forest algorithms were applied, and the random forest model correctly predicted dementia presence with 52% accuracy, whereas the logistic regression model predicted dementia with 48% accuracy. Future work will tune the models to improve the results and determine which factors were most influential in predicting dementia.",
      "research_paper": "assets/pdfs/27b624e4bfd28883a3c6c832af37e5bfee8fd185.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "graphic_link": "https://www.hopehealthco.org/wp-content/uploads/2023/06/iStock-1400657957.jpg",
      "headline": "This research addresses the urgent need for early dementia detection by using machine learning to analyze patient health factors, such as heart rate and diabetes, for dementia prediction, achieving a 52% accuracy with a random forest model and paving the way for future model improvements and deeper insights into influential risk factors.",
      "project_id": "19648840479d377917554319f6f043e4158a67bb",
      "expand": false,
      "tags": [
        "Neuroscience",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        103,
        61,
        58
      ]
    },
    {
      "student_name": "Madhurima M.",
      "mentor_name": "Victoria Lloyd",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/5cf78eaf169dc6acb1daaf5a28e7e73e33d93014.png",
      "domains": [
        "Engineering",
        "Specific Technologies"
      ],
      "project_title": "A Comprehensive Review on Deep Learning Architectures for Image Segmentation",
      "project_desc": "At a micro level, image segmentation is an important computer vision task that assists in applications in various fields such as medical diagnostics, self-driving cars, and robotics. The introduction of deep learning in the scene led to great improvements in the performance of segmentation models in terms of accuracy, efficiency, and size. This paper presents a comprehensive literature review of recent advancements pertaining to the deep learning-based architectures for the image segmentation task. Such models include the well-known U-Net and Mask R-CNN models, as well as more recent models including Attention U-Net, segmentation using Generative Adversarial Networks (GANs) and transformer-based models including Vision Transformer (ViT) for segmentation tasks. We also include the analysis of hybrid architectures that utilize both CNN and transformer components for better performance in segmentation tasks. Other important issues such as data shortage and over-difficulty of annotation tasks are also considered, as well as the influence of data augmentation and few apologies on these problems. In addition to that, we also perform a comparative analysis of these architectures and evaluate and compare their performance on several benchmark datasets including their merits and demerits and usage. Finally, we identify current challenges and suggest promising future research directions which will include the development of more efficient models for real-time deployment and improvements in model interpretability and explainability.",
      "research_paper": "assets/pdfs/2b4f8be15163ac1d35c60b01862b31bcfb08becc.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "graphic_link": "https://www.carierista.com/storage/company-posts/82/266769da4e5df45ec3a5d6947aba04dc.jpg",
      "headline": "This paper provides a comprehensive review of recent advancements in deep learning-based models for image segmentation, including U-Net, Mask R-CNN, and transformer-based models, analyzing their performance, challenges like data shortage, and potential future directions for improved real-time deployment and interpretability.",
      "project_id": "2549cd0399a882042fd1aff5316475dd89af893c",
      "expand": true,
      "tags": [
        "Engineering",
        "Computer Vision"
      ],
      "related_proj": [
        96,
        166,
        144
      ]
    },
    {
      "student_name": "Nikhil T.",
      "mentor_name": "Kyra Kraft",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology",
        "Environment"
      ],
      "project_title": "Using Machine Learning Models to Predict Asthma Hospitalizations Based on Air Pollutants",
      "project_desc": "This paper explores the use of a Decision Tree regression model to predict asthma hospitalizations across 42 regions of New York City based on air pollution data. The study focuses on three key air pollutants: PM2.5, NO2, and O3. The model’s performance is evaluated using Mean Absolute Error, Mean Absolute Percentage Error, R-Squared Score, and Root Mean Squared Error. While the study does not aim to produce definitive forecasts, it assesses the viability of this machine learning approach for identifying health vulnerabilities related to air pollution. Forecasting the asthma hospitalizations by region can help identify and rectify inequities in health vulnerability. The model achieved a Mean Absolute Error of 50.01 and a Mean Absolute Percentage Error of 84.3%, indicating significant prediction challenges. Results show that high levels of NO2 and O3 are associated with increased asthma-related ER visits. These findings highlight potential areas for targeted healthcare interventions.",
      "research_paper": "assets/pdfs/f188fca07a265423cc23b00ec0f6d3a7a9749dd4.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "graphic_link": "https://www.drmayankshukla.com/wp-content/uploads/2023/04/understanding-air-pollution-and-its-effects-on-lungs.jpg",
      "headline": "This study uses a Decision Tree regression model to analyze the relationship between air pollution (PM2.5, NO2, O3) and asthma hospitalizations in New York City, revealing significant prediction challenges and highlighting the potential for targeted healthcare interventions based on regional health vulnerabilities.",
      "project_id": "5c10f4c099ff578f3715eb7c7328dda677ab2433",
      "expand": true,
      "tags": [
        "Pollution",
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        178,
        90,
        33
      ]
    },
    {
      "student_name": "Pracheth G.",
      "mentor_name": "Henry Cerbone",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/edf6eed47f9a15155ef5bfd24aaca257e0a0685f.png",
      "domains": [
        "Environment",
        "Engineering"
      ],
      "project_title": "Smart Agriculture: Optimizing Soil pH and Crop Recommendations Using Machine Learning",
      "project_desc": "Precision agriculture relies heavily on machine learning to provide actionable insights into crop management. This study investigates the application of hyperparameter tuning to improve crop classification models using environmental and soil data. By leveraging a dataset of 22 crop types with features such as nitrogen content, potassium levels, pH, and rainfall, Decision Tree and Random Forest models were developed and optimized. Hyperparameters like max_depth and criterion were systematically adjusted to balance bias and variance. Through extensive tuning, the Decision Tree Classifier achieved a classification accuracy of 99.5%, while the Random Forest Regressor demonstrated a Mean Absolute Error (MAE) of less than 0.4. This paper explores the impact of hyperparameter optimization on model performance and highlights its practical implications for improving crop recommendation systems in agriculture. The findings demonstrate the value of AI-driven decision-making in sustainable farming and pave the way for future innovations in agricultural technology.",
      "research_paper": "assets/pdfs/4497be77c60dd8fe76cace0249cbb99258a071e5.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "graphic_link": "https://www.innominds.com/hubfs/Innominds-201612/img/IM-News-and-Blogs/Innominds-Solution-for-Precision-Agriculture.jpg",
      "headline": "This study explores how hyperparameter tuning enhances the performance of machine learning models, achieving a 99.5% accuracy in crop classification and highlighting the potential of AI-driven optimization for sustainable precision agriculture.",
      "project_id": "9f183f967fec1c178bcd92ceb8019b191e137564",
      "expand": true,
      "tags": [
        "Environmentalism",
        "Farming",
        "Irrigation"
      ],
      "related_proj": [
        161,
        120,
        30
      ]
    },
    {
      "student_name": "Paul N.",
      "mentor_name": "Erick Ruiz",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/901fc534a3310b19a0a26835fab35feedcf61a51.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Using Machine Learning for Calculus",
      "project_desc": "The goal of this research project is to develop a machine learning program which can take an integral equality (for example, a u-substitution that turns one integral into another) and verify that it is correct. Many other mathematics-related software can only take in a one-sided equation and return a calculated output, but most mathematical proofs are built on relating several successive pieces of logic. For this problem, a language processing approach was taken to convert the integral equations into something that could be understood by a language processing model. From there, several different classifiers were tested, each one returning about a 40-60% accuracy depending on the dataset that was input.",
      "research_paper": "assets/pdfs/6b22ba84081780b97f1e1e6942ddd411e43cb3eb.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "graphic_link": "https://online.stanford.edu/sites/default/files/styles/embedded_large/public/2018-09/computer-science-continuous-mathematical-methods-machine-learning_CS205L.jpg?itok=QzW1BY2F",
      "headline": "This research aims to develop a machine learning program that verifies the correctness of integral equalities by leveraging a language processing approach, achieving 40-60% accuracy across different classifiers and datasets, unlike traditional mathematical software that only evaluates one-sided equations.",
      "project_id": "3756d55acae0d7d4f98b9774a09702adf87a4f1b",
      "expand": false,
      "tags": [
        "Engineering"
      ],
      "related_proj": [
        30,
        167,
        181
      ]
    },
    {
      "student_name": "Advait K.",
      "mentor_name": "Abdulla Kerimov",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/c9b7628e4f688209721d1fa569c6570eccd85dca.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Early Detection of Knee Arthritis Using AI: Enhancing Diagnostic Accuracy with Deep Learning",
      "project_desc": "Arthritis, particularly of the knee, is one of the most common chronic diseases worldwide and seriously hinders patients' mobility and quality of life. Early diagnosis is essential for effective intervention; however, diagnosis using X-ray imaging is difficult due to the subtlety and variability manifested in the various states of the disease. Deep learning methods, in particular, convolutional neural networks (CNNs), are proposed for the automatic grading of knee arthritis into five classes: Normal, Doubtful, Mild, Moderate, and Severe. This is an artificial intelligence system that avoids the many difficulties of manual diagnosis, which are inconsistent between radiologists and very time-consuming with current methods, to increase the diagnostic accuracy and consistency, especially in developing or resource-constrained settings with limited specialist access.\n\nThe study uses transfer learning with state-of-the-art models, such as EfficientNetB0, DenseNet121, and InceptionV3, to improve the classification of arthritis severity on X-ray images. Pre-trained layers on ImageNet datasets were fine-tuned for the knee arthritis dataset by adding custom classification heads to each model. Data preprocessing methods, including resizing and normalization, were also applied to ensure uniformity throughout the dataset. Their performances were tested with the use of such metrics as accuracy, precision, recall, and F1-score, validated against different stages in the development of arthritis. Of special note was EfficientNetB0, showing better results of up to an 85% accuracy score for this overall scenario.\n\nWhile the models showed promising results, they still struggled with the accurate classification of cases of Mild arthritis, indicating a need for further refinement. This study has brought out the potential of AI-based diagnostic tools in reducing diagnostic variability, speeding up the determination of arthritis severity, and improving patient outcomes. However, it also responds to key concerns about the \"black box\" nature of deep learning models, which can impede clinical adoption. To tackle this challenge, the research highlights the importance of explainable AI methodologies in building trust in practitioners of healthcare through the interpretability of predictive models.\n\nThis study has significant implications for the medical field, as it demonstrates the potential of AI in helping radiologists and primary care providers achieve better accuracy and consistency in diagnosing knee arthritis. Their deployment in mobile or cloud-based settings may greatly increase the access to diagnostic tools in remote or disadvantaged areas. Future efforts in research will go in the direction of increasing diversity in datasets, adding multimodal data, and interpretability of models to encourage widespread adoption in clinical practice. To make effective and scalable artificial intelligence solutions available for more precise, rapid, and fairer diagnoses of knee arthritis, with an overall goal to have better patient management and care in mind.",
      "research_paper": "assets/pdfs/27d64189cc8d9ca550e9cf78ad9f7858c2495ff2.pdf",
      "project_yr": "2024",
      "project_quarter": "Winter",
      "github": "https://github.com/AdvaitKothuri/Knee-Arthiritis-Detection/tree/main",
      "graphic_link": "https://media.istockphoto.com/id/1213394222/photo/joint-pain-arthritis-and-tendon-problems-a-man-touching-nee-at-pain-point-on-white-background.jpg?s=612x612&w=0&k=20&c=uisKEzXJXLJYHjiTYEOL-LNp-IQ_6ekK-e-Fy_wiRAg=",
      "headline": "This research explores the use of deep learning models, specifically convolutional neural networks like EfficientNetB0, DenseNet121, and InceptionV3, to improve the accuracy, consistency, and accessibility of knee arthritis diagnosis through automated classification of disease severity, while emphasizing the need for explainable AI to enhance clinical trust and adoption.",
      "project_id": "7db4578136d1f58aa125d3621a5787aea4398796",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions"
      ],
      "related_proj": [
        152,
        178,
        29
      ]
    },
    {
      "student_name": "Brayden T.",
      "mentor_name": "Rima Rebei",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/ab03fd7689c202a7dde6a94e4885fb41de6e211b.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Path Planning for Autonomous Cinematic Drones",
      "project_desc": "This project explores the autonomous movement of drones in aerial cinematography. Specifically, based on certain locations in a scene, I examined algorithms that will allow a drone to autonomously find the smoothest path to reach each location while maintaining smooth camera quality. Drones play a significant role in modern cinematography, particularly in creating dynamic aerial shots and panoramic photographs. Given their importance, drones must operate smoothly and reliably. Implementing autonomous path planning will reduce common challenges, such as collision risks, while also limiting manual navigation, minimizing the need for human intervention while enabling safer, more efficient, and faster filming with high-definition (HD) video quality. I used a simulation system, Gazebo, and the ROS (Robot Operating System) interface as an open-source platform for coding, and together they simulated autonomous drone movements. By creating and using multiple path planning algorithms—A-star (A*), Dijkstra’s, and Greedy Best First Search (GBFS)—I enabled the drone to autonomously move to a certain position while accounting for obstacles in a static environment. The drones were able to efficiently detect obstacles and calculate optimal routes, with A* demonstrating the highest average efficiency. In comparison to Dijkstra’s algorithm, A* often reduced pathfinding time by around 70%. Additionally, its paths were generally more accurate than the GBFS algorithm in various situations, reducing travel time by around 30% on average when larger obstacles and environments were tested. The results of this study present A* as the optimal path planning algorithm that can be implemented to improve autonomous drone cinematography in future applications. Furthermore, another form of path planning, which uses roadmaps to autonomously move a drone, is also explored to help prevent concerns with restricted areas and cluttered environments, and to allow personalized cinematography on human-made paths.",
      "research_paper": "assets/pdfs/cf7c938bf3394d17b99bf93912c7d5475fa170ff.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://cdn.fstoppers.com/styles/large-16-9/s3/lead/2023/12/nando-drone-header.jpg",
      "headline": "This study explores autonomous drone navigation for aerial cinematography, focusing on algorithms that enable drones to smoothly reach specified locations while maintaining high-definition camera quality. Using a simulation environment, the study tested multiple path planning algorithms and found that A* consistently outperformed others in both accuracy and efficiency, making it the most suitable for enhancing autonomous drone cinematography.",
      "project_id": "3b5df0df74a3d01151a7b76374a074efe5554e28",
      "expand": false,
      "tags": [
        "Engineering",
        "Robotics"
      ],
      "related_proj": [
        170,
        159,
        172
      ]
    },
    {
      "student_name": "Ananya S.",
      "mentor_name": "Nancy Zhu",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/23c6962a2d4b900a6d4b9541f26c136a73a7e5ba.png",
      "domains": [
        "Engineering",
        "Arts and Music"
      ],
      "project_title": "Classifying Carnatic Ragams Based On Audio Features",
      "project_desc": "Carnatic music is a form of classical singing originating from South India. It consists of many nuances and patterns that are organized into various ragams, which are melodic frameworks for a song that dictate the types of notes and modulations present. Each ragam contains complicated variations of notes and features, however, so it is often difficult for singers and listeners to identify ragams. To address this challenge, I built a model using a Random Forest Classifier that analyzes features such as Chroma Short-Time Fourier Transform, Spectral Centroid, Spectral Bandwidth, various MFCC coefficients, and the ragam label in order to identify the ragam of the audio clip. The model was trained on a dataset of 1112 audio clips that includes 32 ragams– 7 melakarta “parent” ragams and 25 janya “child” ragams (Sanjana Satish68l). Each ragam has certain patterns within its features that make it easy to distinguish from other ragams, and therefore easy for the model to categorize. By analyzing these distinct features, the model was able to accurately classify the ragam of almost all the test clips. The accuracy was considerably high: around 99.1%, which outperformed both the K Nearest Neighbors and Decision Tree models. ",
      "research_paper": "assets/pdfs/b8e5ba604e4629d9a7b4085c8357293e5efb0b9c.pdf",
      "project_yr": "2025",
      "project_quarter": "Spring",
      "graphic_link": "https://images.news18.com/ibnlive/uploads/2021/07/1627729052_carnaticmusic-1600x900.png",
      "headline": "This study developed a Random Forest Classifier to identify ragams in Carnatic music by analyzing key audio features such as Chroma STFT, spectral properties, and MFCC coefficients. Trained on a dataset of 1,112 audio clips across 32 ragams, the model achieved a high classification accuracy of 99.1%, outperforming other baseline models.",
      "project_id": "3909d61520a0745e03e226823de080a4a5d7bb70",
      "expand": false,
      "tags": [
        "Engineering",
        "Music"
      ],
      "related_proj": [
        30,
        141,
        161
      ]
    },
    {
      "student_name": "Jia A.",
      "mentor_name": "Shreyaa Raghavan",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/3fdaa8acfab47133d6ad5db412af8cd3bc07e427.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Predicting Investment Success on Shark Tank India: A Machine Learning Approach ",
      "project_desc": "Shark Tank is a critical platform for emerging entrepreneurs, offering not only financial investment but also valuable visibility from renowned investors, or \"Sharks.\" While extensive research has been conducted on Shark Tank in Western markets, there is a notable gap in studies focusing on the Indian context. India has one of the fastest-growing economies in the world, with a rapidly expanding startup ecosystem that is significantly different from Western markets. Additionally, Indian entrepreneurs face unique challenges, such as limited access to funding, regional disparities, and a complex regulatory environment, which make the Shark Tank platform particularly impactful for their success. Understanding the specific factors that influence investment decisions in this context—such as investor preferences, market fit, and demographic considerations—can provide invaluable insights for entrepreneurs looking to secure funding. This study addresses the challenge of predicting whether a pitch will receive an offer by leveraging a diverse dataset with input factors from both U.S. and Indian datasets. It employs a range of regression models, a neural network, and transfer learning techniques to adapt insights from U.S. data to the Indian context, improving the model's predictive accuracy across different markets.    Results indicate that the neural network model achieved the highest predictive accuracy, capturing complex interactions better than simpler models. Remarkably, transfer learning provided reasonable results even after removing city and state features, underscoring its adaptability across diverse markets. These findings highlight the potential of advanced machine learning techniques to improve understanding of investment dynamics in emerging economies like India.",
      "research_paper": "assets/pdfs/5d27a9d3229f74b3316c5f22ed386193016ba19a.pdf",
      "project_yr": "2024",
      "project_quarter": "Fall",
      "graphic_link": "https://techcrunch.com/wp-content/uploads/2023/07/sti_s2_set6_5may_landscape_thumb.jpg",
      "headline": "This project explores the use of machine learning—specifically regression models, neural networks, and transfer learning—to predict the success of startup pitches on Shark Tank, with a focus on bridging insights between U.S. and Indian markets. By addressing the unique challenges faced by Indian entrepreneurs, the project demonstrates how advanced modeling can uncover patterns in investment decisions and support funding opportunities in emerging economies.",
      "project_id": "97d5556bd58ffcb6b344c3369e1d8de45960cf6c",
      "expand": true,
      "tags": [
        "Consumer Behavior",
        "Pricing",
        "Economics"
      ],
      "related_proj": [
        46,
        88,
        186
      ]
    },
    {
      "student_name": "Adi M.",
      "mentor_name": "Abdullah Rashed Ahmed",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/3fcb658eaad1ae61a87279a102133a78bf2e1464.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Diabetic Retinopathy Evaluation: A ViT Approach",
      "project_desc": "Diabetic Retinopathy (DR) is a complication of diabetes, which afflicts about 400 million people in today's world, that causes the blood vessels of the retina to swell and to leak fluids and blood. Early detection of DR can help manage it with lifestyle changes and medical interventions, preventing the need for costly treatments like laser or surgery, and providing a much better standard of life. This motivates the search for easier, affordable and accessible DR detection technology using the help of Artificial Intelligence given the huge number of potential patients.\n\nThis paper investigates the application of a Vision Transformer (ViT) model for classifying diabetic retinopathy severity levels from fundus images, aiming to provide accurate predictions essential for early diagnosis and treatment.\n\nTo optimize model performance, images were preprocessed to appropriate resolutions, and the dataset was divided into training, validation, and test sets. Data augmentation techniques, including geometric transformations and color adjustments, were employed to enhance data diversity and prevent overfitting. The ViT model was modified by freezing upper layers and adjusting the final fully connected layer to classify images across five severity levels, leveraging transfer learning for efficient fine-tuning.\n\nWith a learning rate of 0.0002 and a batch size of 52, the model was trained with early stopping, achieving a validation accuracy of 75.96%, a loss of 0.58, a recall of 0.72, a precision of 0.700, and an F1-score of 0.675.These results demonstrate the model’s balanced performance, capturing most positive cases while maintaining reasonable precision. Yet, the results fell short of 90%+ accuracy obtained in [8] throughout our testing period and more analysis is necessary to assess. Future work will explore dynamic learning rate schedules and alternative batch sizes to further enhance model accuracy to reach the higher goal.\n\nThis study highlights the Vision Transformer model’s potential for nuanced medical image classification tasks, contributing to advancements in automated diagnostic tools.",
      "research_paper": "assets/pdfs/7c3d8127d86f37f0c151ae5e435eb7ffe5407958.pdf",
      "project_yr": "2025",
      "project_quarter": "Spring",
      "github": "https://github.com/ManjuDaBest/ViT-for-Diabetic-Retinopathy",
      "graphic_link": "https://media.licdn.com/dms/image/v2/C4E12AQEgZI8jM5lHGw/article-cover_image-shrink_720_1280/article-cover_image-shrink_720_1280/0/1520128677512?e=2147483647&v=beta&t=UJWJxBuwZsLpi2A15aYo0Xp3lPwJb1cQKpNNWP3Ltqk",
      "headline": "This study explores the use of a Vision Transformer (ViT) model to classify diabetic retinopathy severity from fundus images, demonstrating promising accuracy through transfer learning and data augmentation techniques. While results fell short of state-of-the-art benchmarks, the model shows potential for improving early diagnosis through AI-driven medical imaging.",
      "project_id": "b779de4df863905eac36a31c3fc8c0cf7baadec9",
      "expand": false,
      "tags": [
        "Diseases and Health Conditions",
        "Eyes"
      ],
      "related_proj": [
        143,
        125,
        157
      ]
    },
    {
      "student_name": "Satvi M.",
      "mentor_name": "Rebecca Funke",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/c2721de7183219cf2f1d980ccafd729877e592f5.png",
      "domains": [
        "Generative AI"
      ],
      "project_title": "Art Intel - Making AI your Ally in the Creative Process",
      "project_desc": "“Artificial Intelligence is demolishing our creativity.”  These words have been uttered by nearly every artist today. People use AI to create art, but not in the ethical way that it should be used, that is, the art generated by the AI is taken and claimed as their original work. This is often the result of a lack of creativity, but it can also cause creativity to reduce as well. So, instead of using AI in a way that hinders the creative process, I wanted to use it in a way that promotes the process. My app Art Intel combines my passion for art and my interest in artificial intelligence by utilizing generative AI to serve as an ally in the creative process to solve that problem. Rather than the AI giving back an image with the fixes made, making it easy for the artist to steal it and claim it as their own, Art Intel will be by your side to guide you down the path from good art to great art, without taking away creative liberties.",
      "research_paper": "assets/pdfs/ddb3d9503e3dd25d13972fa01442c2910d6b037b.pdf",
      "project_yr": "2025",
      "project_quarter": "Spring",
      "graphic_link": "https://www.adobe.com/products/firefly/features/media_179810889bf1ef34a453137e0387dd9e0f4e43f05.jpg?width=750&format=jpg&optimize=medium",
      "headline": "Art Intel is an app that uses generative AI to support and enhance the artistic process without replacing or undermining human creativity. Instead of producing finished pieces for users to claim, it guides artists in refining their work while preserving their creative control.",
      "project_id": "1906b2258ade3dafbdcc51f2105a37d662969b4b",
      "expand": false,
      "tags": [
        "Visual Arts",
        "Generative AI"
      ],
      "related_proj": [
        162,
        76,
        173
      ]
    },
    {
      "student_name": "Quang M.",
      "mentor_name": "Edoardo Contente",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/2ac6f291b540d7d9691e3968998b278d3d25c51d.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Evaluating Autonomous Vehicle Safety: A Comparative Analysis of ADS and ADAS Incident Data Using Machine Learning Models",
      "project_desc": "This paper focuses on Autonomous Vehicles (AVs), which are rising rapidly in road transportation.\nHowever, the reliability of AV systems, namely Advanced Driver Assistance Systems (ADAS) and Au-\ntomated Driving Systems (ADS), remain vague. Therefore, it is essential that we improve their safety\nmeasurements. This study aims to evaluate and compare the safety performance of ADAS and ADS by\ntraining supervised machine learning models on incident report data, specifically focusing on predicting\ninjury severity, using numerical and categorical data related to crash conditions and environmental fac-\ntors. We compared different models, for each fine-tuning their respective hyperparameters. The best per-\nformance was achieve with Random Forest with hyperparameters max depth = 20, n estimators = 200.\nThis work is expected to provide valuable insights for both the improvement of autonomous systems and\noverall road safety",
      "research_paper": "assets/pdfs/0a047689ffe3b374aa71213277c10d4d29d39b44.pdf",
      "project_yr": "2024",
      "project_quarter": "Summer",
      "graphic_link": "https://media.wired.com/photos/65bd5ee7960589665dfbad00/master/pass/AutonomousSystems-bySenorSalme-Lede-Quanta-Science.jpg",
      "headline": "This research evaluates the safety performance of Advanced Driver Assistance Systems (ADAS) and Automated Driving Systems (ADS) by training supervised machine learning models on incident report data to predict injury severity. The best results were achieved using a fine-tuned Random Forest model, offering insights to improve autonomous systems and enhance road safety.",
      "project_id": "e85f46fe40645bc97eb2bb933d8c877951e70d60",
      "expand": true,
      "tags": [
        "Autonomous Vehicles",
        "Engineering"
      ],
      "related_proj": [
        144,
        52,
        110
      ]
    },
    {
      "student_name": "Pandian A.",
      "mentor_name": "Udgam Goyal",
      "mentor_title": "MEng CS from MIT, Product Manager at Aurora and prior AI Product Management at Microsoft",
      "mentor_image": "assets/images/mentor_imgs/a3638f2f762e49748ac4a432468a44b6f5a7457b.png",
      "student_image": "assets/images/student_imgs/79302f756a3a58d4cd41313a5b4080e394fc38be.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Enhancement of Autonomous Vehicles During Extreme Weather",
      "project_desc": "Autonomous vehicles are self-driving vehicles that operate without human assistance. Self-driving vehicles, which include Teslas, BMWs, Waymos, Fords, and Mercedes, struggle to detect objects under extreme weather conditions, such as rain and fog, due to a lack of visibility, thereby significantly increasing the likelihood of collisions. Although autonomous vehicles have proven to be exceptional in nominal conditions, their poor performance in stormy, foggy, and snowy conditions currently limits their deployment on real roads today. This is important to know because AI can be used to improve the safety of vehicles using algorithms like deep learning via neural networks. In order to answer the question, it involved training and fine-tuning a neural network called the YOLOv8n  model. We loaded a pretrained model, called yolov8n. This model was inaccurate at predicting cars when the vision isn’t clear. However, from the DAWN dataset from Kaggle + YOLOV8, we fine-tuned the model and got it to predict vehicles when the vision wasn’t clear. Initially, the model failed to predict other vehicles in extreme weather conditions, or often predicted other vehicles and objects with low confidence levels. However, after the model was fine-tuned, it was able to increase its prediction confidence level by 40%, also being able to predict 91% of the presented objects, way higher compared to its predicting 22% of its presented objects per image. However, during normal conditions, the fine-tuned model predicted other vehicles at a lower confidence level compared to the original Yolov8n model due to it being trained during extreme weather and not nominal conditions. It was concluded that the fine-tuning of YOLOv8n contributed to the feasibility of object detection during extreme weather conditions, with a 64% increase in objects detected with it being able to detect 94.87% of vehicles and pedestrians,  but led to a 10% decrease in confidence level and <1% decrease in object detection rate under nominal conditions and failed to detect traffic lights, which the non-tuned model could do. This makes it necessary for self-driving providers to build a car with at least two different models: one non-tuned YOLOv8n model, operating during nominal conditions and during traffic lights, and the other being a tuned YOLOv8n model that operates during extreme weather. ",
      "research_paper": "assets/pdfs/99f8fd9c50c25bf80336e82eec3e06099673fb0d.pdf",
      "project_yr": "2025",
      "project_quarter": "Fall",
      "github": "https://github.com/PandianAnand/Autonomous-Vehicle-Improvement",
      "graphic_link": "https://cms.accuweather.com/wp-content/uploads/2016/10/650x366_10101621_650x366_10080011_fordnaias_autonomy_05.jpg",
      "headline": "This research investigates improving object detection for autonomous vehicles in extreme weather by fine-tuning the YOLOv8n neural network. Starting with a pretrained model that struggled to recognize vehicles in rain, fog, and snow, we used the DAWN dataset from Kaggle to retrain it, raising detection accuracy from about 22 percent to over 90 percent in poor-visibility conditions while increasing confidence levels by roughly 40 percent, though at the cost of slightly reduced performance and traffic-light detection in normal conditions.",
      "project_id": "2fcfe3ae3e0ac0b0bd88127727f8578cb517626d",
      "expand": false,
      "tags": [
        "Autonomous Vehicles",
        "Engineering"
      ],
      "related_proj": [
        99,
        140,
        163
      ]
    },
    {
      "student_name": "Richa K.",
      "mentor_name": "Rami Abi-Akl",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/c100ee99327ea9144c621d12294741d2dc993dbc.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Predicting Unplanned Electric Vehicle Breakdowns using Machine Learning on Sensor and Diagnostic Data",
      "project_desc": "Unplanned breakdowns in electric vehicles (EVs) pose challenges such as costly repairs, unexpected downtime, and reduced user confidence. To address this issue, this research investigates whether artificial intelligence (AI) can perform fault prediction and improve EV reliability. Using the EV Sensors, Driving Pattern & Diagnostics dataset (2020–2024), which contains information on battery performance, motor readings, driving behaviors, and diagnostic trouble codes (DTCs), we tested four machine learning models: Random Forest, Decision Tree, Support Vector Regression (SVR), and Linear Regression. Results showed that the Decision Tree and Random Forest models outperformed the others, achieving over 90% predictive accuracy compared to less than 70% for SVR and Linear Regression. These findings indicate that tree-based model approaches are effective in capturing nonlinear patterns between sensor data and breakdown events. Overall, this study highlights the potential of AI to provide early warnings of EV failures, enable proactive maintenance strategies, and strengthen consumer trust in EV technology.",
      "research_paper": "assets/pdfs/685ea632d2a05cedff92776d0529981ffa0317fa.pdf",
      "project_yr": "2025",
      "project_quarter": "Fall",
      "graphic_link": "https://www.icsense.com/wp-content/uploads/2017/09/iStock-638796582-1.jpg",
      "headline": "This study explores the use of artificial intelligence (AI) for fault prediction in electric vehicles (EVs) to reduce unexpected breakdowns and improve reliability. Using sensor, driving, and diagnostic data from 2020–2024, the research found that tree-based models, particularly Decision Tree and Random Forest, achieved over 90% predictive accuracy, demonstrating AI’s potential to enable proactive maintenance and enhance user confidence in EV technology.",
      "project_id": "729ac5a4b9a5e1389a85eb5fa4324202cc752bf2",
      "expand": true,
      "tags": [
        "Autonomous Vehicles",
        "Optimization",
        "Engineering"
      ],
      "related_proj": [
        166,
        173,
        144
      ]
    },
    {
      "student_name": "Arya S.",
      "mentor_name": "Khanh Nguyen",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/c0bde7e1a1563651260661c990c2c2b880c76034.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Structure-Based Analysis of Parkinson’s Disease Protein Mutations:  Comparative Analysis of Computational and Experimental SNCA Structures",
      "project_desc": "Parkinson’s disease (PD) is a progressive neurodegenerative disorder strongly associated with the misfolding and aggregation of α-synuclein (SNCA). Familial mutations such as A30P, A53T, and E46K are known to accelerate aggregation and alter protein–ligand interactions, but structural data remain limited due to the intrinsically disordered nature of SNCA. This presents a major challenge for therapeutic design. Recent advances in artificial intelligence, including AlphaFold and ColabFold, enable the prediction of protein structures directly from sequence. While these tools have transformed structural biology, their reliability for intrinsically disordered proteins like SNCA remains uncertain. In this study, we compare computationally predicted wild-type and mutant SNCA models against experimentally determined structures, using molecular docking and binding affinity prediction with pharmacologically relevant ligands. By integrating protein structure prediction with ligand docking, this work highlights both the potential and the limitations of AI-based methods for studying mutation-specific effects in PD. More broadly, it demonstrates how computational modeling can complement experimental approaches to accelerate drug discovery efforts targeting α-synuclein pathology.",
      "research_paper": "assets/pdfs/00b315fa097dc0bf4293d3bad27aa6cfc60c6e01.pdf",
      "project_yr": "2025",
      "project_quarter": "Fall",
      "graphic_link": "https://assets.technologynetworks.com/production/dynamic/images/content/392556/large-scale-map-tracks-how-mutations-misguide-proteins-in-cells-392556-960x540.jpg?cb=13057049",
      "headline": "This study uses AI-based protein modeling tools such as AlphaFold and ColabFold to compare wild-type and mutant α-synuclein structures in Parkinson’s disease, evaluating their interactions with pharmacologically relevant ligands. By integrating structure prediction with molecular docking, the research highlights both the potential and limitations of computational methods for understanding mutation-specific effects and guiding drug discovery efforts.",
      "project_id": "36f0fafbfef808ac4301a663f4a3135b18cfdf5a",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Neuroscience"
      ],
      "related_proj": [
        93,
        117,
        101
      ]
    },
    {
      "student_name": "Samuel P.",
      "mentor_name": "Rami Abi-Akl",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Using Machine Learning Models to Predict Heart Disease and Attack with Non-Clinical and Non-Wearable Variables",
      "project_desc": "Heart disease is the leading cause of death in the United States for over 100 years. However, electrocardiograms (EKG), the primary method of testing for heart issues, costs the average American ~100-5000 dollars per scan. Therefore, there should be other cheaper and time efficient methods for other methods of early onset heart disease detection. We used data from the  Behavioral Risk Factor Surveillance System (BRFSS), a telephoned based survey with over 200 thousand participants and partnered with theCenters for Disease Control and Prevention (CDC), to find correlations between heart disease and attacks with non-clinical factors. Multiple machine learning models were trained on BRFSS 2015 data. The best model achieved 70% accuracy, while having a 21% precision and a 79% recall. These results show that while there is potential using non-clinical data combined with machine learning to provide a cost-effective preliminary screening tool for heart disease risk, significant improvements are needed before they can be applied effectively in real-world healthcare.",
      "research_paper": "assets/pdfs/1c65875e316d359f138260fc8f551e521516c38d.pdf",
      "project_yr": "2025",
      "project_quarter": "Fall",
      "graphic_link": "https://scx2.b-cdn.net/gfx/news/2025/engineers-develop-wear.jpg",
      "headline": "This study uses data from the CDC’s Behavioral Risk Factor Surveillance System (BRFSS) to train machine learning models that predict heart disease risk based on non-clinical factors. The research demonstrates the potential for cost-effective, data-driven screening tools for early heart disease detection, though further improvements are needed for real-world application.",
      "project_id": "799b3a74ad6e3f41dbf3855f60f38a3ab78c7272",
      "expand": true,
      "tags": [
        "Diseases and Health Conditions",
        "Heart Disease"
      ],
      "related_proj": [
        171,
        148,
        24
      ]
    },
    {
      "student_name": "Nimeesha K.",
      "mentor_name": "Serena Hughes",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/missing_image.png",
      "domains": [
        "Healthcare and Biology"
      ],
      "project_title": "Modeling Sleep Efficiency",
      "project_desc": "I’ve always been interested in sleep and why some people tend to sleep better than others. I think it’s important to recognize what could be affecting a person’s sleep and what they can do to improve it. The goal of my research question, “How much do alcohol, exercise, and smoking affect sleep?” is to show how factors such as these can affect the quality of one’s sleep. I found a data set consisting of many factors that can affect sleep, as well as sleep quality. Using this data set, I built a model that allows you to input values for factors and come out with a sleep quality score. My model is a regression model, as it makes predictions through numerical data given to it. I tested out different models, settling on the most accurate one, Linear Regression, with a mean absolute error of 0.051. ",
      "research_paper": "assets/pdfs/85be12d9d861f4964683a7f7bd125252d174c1e2.pdf",
      "project_yr": "2023",
      "project_quarter": "Summer",
      "graphic_link": "https://img.freepik.com/free-vector/hand-drawn-bedtime-stories-illustration_23-2151162751.jpg?semt=ais_hybrid&w=740&q=80",
      "headline": "This study explores how alcohol use, exercise, and smoking affect sleep quality by analyzing a dataset containing various lifestyle and sleep-related factors. Using a regression-based predictive model, the research identifies relationships between these behaviors and sleep outcomes, with the final linear regression model achieving a mean absolute error of 0.051.",
      "project_id": "59f879b17f79dcb9791157e12ba7cd82e08be230",
      "expand": false,
      "tags": [
        "Physiology"
      ],
      "related_proj": [
        7,
        48,
        178
      ]
    },
    {
      "student_name": "Rishi N.",
      "mentor_name": "John Basbagill",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/a8845545754fe4c102275360d00278c79f673263.png",
      "domains": [
        "Business and Finance"
      ],
      "project_title": "Forecasting Bitcoin Price Direction from Twitter Sentiment and Market Data",
      "project_desc": "Forecasting Bitcoin Price Direction from Twitter Sentiment and Market Data, explores how social media emotions influence cryptocurrency prices. Using the 17.7 million Bitcoin Tweets dataset from Kaggle (2017–2019), I  applied VADER sentiment analysis to link tweet tone with hourly market data. After cleaning, visualizing, and preprocessing the data, I trained two machine learning models—Random Forest and XGBoost—to predict Bitcoin’s price direction. XGBoost performed best, reaching 74.15% accuracy and showing that both market and sentiment features (like Compound Score and tweet volume) were key predictors. Future plans include feature engineering, hyperparameter tuning, and combining models for improved accuracy.",
      "research_paper": "assets/pdfs/fd15d96cc3e54b9d19e963f7bba7b85e8ab4cb61.pdf",
      "project_yr": "2025",
      "project_quarter": "Fall",
      "graphic_link": "https://m.economictimes.com/thumb/msid-121432978,width-1600,height-900,resizemode-4,imgsize-159732/crypto-prices-today-bitcoin-slips-below-110000-amid-profit-booking-altcoins-fall-up-to-5.jpg",
      "headline": "This research explores how social media sentiment affects cryptocurrency price movements using the 17.7 million Bitcoin Tweets dataset from Kaggle. By applying VADER sentiment analysis and training Random Forest and XGBoost models, the study achieved 74.15% accuracy, highlighting sentiment and market features as key predictors.",
      "project_id": "d117b46e2f6c8ed2887aaf51127c2b75a42b2a11",
      "expand": false,
      "tags": [
        "Cryptocurrency",
        "Economics",
        "Stocks"
      ],
      "related_proj": [
        62,
        122,
        88
      ]
    },
    {
      "student_name": "Humayl F.",
      "mentor_name": "Rami Abi-Akl",
      "mentor_image": "assets/images/missing_image.png",
      "student_image": "assets/images/student_imgs/8a7b0b743c14eb2cd6b7f993e3b2923256bed84a.png",
      "domains": [
        "Engineering"
      ],
      "project_title": "Adapting Martian Rover Technology for Terrestrial Disaster Response and Extreme Terrains",
      "project_desc": "Rovers have transformed planetary exploration by enabling autonomous, intelligent operation on alien terrains. From the Soviet Lunokhod 1 to NASA’s Perseverance, rover technology has evolved to include advanced mobility systems, AI-driven navigation, autonomous decision-making, and onboard scientific instruments. This paper examines how these innovations, originally developed for Mars, can be adapted to solve real-world challenges on Earth, particularly in disaster-prone and inaccessible regions. We designed and 3D-printed a rover prototype, informed by Martian mobility principles, and analyzed its potential for navigating extreme terrain. Using global disaster datasets and a review of current aerial and terrestrial solutions, we evaluated rover applicability to scenarios such as floods, landslides, wildfires, and harsh geographical regions. Our results indicate that ground-based rovers, leveraging autonomous navigation and robust suspension systems, could complement drones and helicopters to improve disaster response and accessibility. Future work will explore multi-agent coordination, hybrid aerial-ground systems, and adaptive AI algorithms to enhance rover efficiency and resilience in complex Earth environments.",
      "research_paper": "assets/pdfs/14965d718ce917780c478def3a984a6db7da129b.pdf",
      "project_yr": "2025",
      "project_quarter": "Fall",
      "graphic_link": "https://cdna.artstation.com/p/assets/images/images/029/001/340/4k/alex-nice-alex-nice-rover-sunset-v011-cameraview.jpg?1596148525",
      "headline": "This research examines how technological advances in Martian rovers, from autonomous navigation to adaptive suspension systems, can inform solutions for disaster response and operation in inaccessible regions on Earth. A 3D-printed rover prototype and global disaster data were used to evaluate the potential of ground-based rovers to complement aerial systems in challenging environments.",
      "project_id": "8873d2f28e0daac08541785f3440c579a7dc456e",
      "expand": true,
      "tags": [
        "Robotics",
        "Hardware",
        "Optimization"
      ],
      "related_proj": [
        170,
        52,
        144
      ]
    }
  ],
  "topics": [
    "Business and Finance",
    "Education",
    "Engineering",
    "Environment",
    "Generative AI",
    "Healthcare and Biology",
    "Humanitarian Efforts and Social Justice",
    "Journalism",
    "Language and Literature",
    "Misc",
    "Physics",
    "PoliSci and Law",
    "Psychology",
    "Published Papers and Science Fairs",
    "Specific Technologies",
    "Sports and Motion Sciences"
  ],
  "tags": [
    "Aerospace Engineering",
    "Alzheimer's",
    "App Development",
    "Astronomy",
    "Audio Analysis",
    "Autonomous Vehicles",
    "BERT",
    "Baseball",
    "Biology",
    "Biotechnology",
    "Brain Cancer",
    "CNNs",
    "COVID-19",
    "Cancer",
    "Chemicals",
    "Chemistry",
    "Classification",
    "Climate Change",
    "Computer Vision",
    "Consumer Behavior",
    "Criminal Justice",
    "Cryptocurrency",
    "DNA",
    "Data and Privacy",
    "Deep Learning",
    "Dental Health",
    "Depression",
    "Developmental Disorders",
    "Diabetes",
    "Diseases",
    "Diseases and Health Conditions",
    "Dropout Rates",
    "Drug Overdose",
    "Drugs",
    "E-Commerce",
    "EOL",
    "Earth Systems",
    "Economics",
    "Education",
    "Engineering",
    "Environmentalism",
    "Eyes",
    "Facial Emotion",
    "Fake News",
    "Farming",
    "Fencing",
    "Food and Beverage",
    "Forest Fires",
    "Fraud",
    "GHGs",
    "GPT-3",
    "Gamma Rays",
    "Gender Equality",
    "Generative AI",
    "Google",
    "Government Systems",
    "Gradient Boosting",
    "Hardware",
    "Heart Disease",
    "Inequality",
    "Irrigation",
    "Kinematics",
    "LSTMs",
    "Law and Order",
    "Leukemia",
    "Linguistics",
    "Logistic Regression",
    "Lung Cancer",
    "Mental Health",
    "Microfinance",
    "Modern Languages",
    "Music",
    "Natural Disasters",
    "Natural Language Processing",
    "Neuroscience",
    "New York",
    "News",
    "Object Detection Models",
    "Optimization",
    "Physics",
    "Physiology",
    "Plants",
    "Pneumonia",
    "Political Science",
    "Pollution",
    "Pop Culture",
    "Prediction",
    "Pricing",
    "Product Analysis",
    "Psych",
    "Published",
    "Racial Equality",
    "Random Forests",
    "Real Estate",
    "Recidivism",
    "Recycling",
    "Reinforcement Learning",
    "Renewable Energy",
    "Robotics",
    "Running",
    "Satellite Imagery",
    "Science Fairs",
    "Skin Cancer",
    "Social Justice",
    "Social Media",
    "Sociology",
    "Solar Cycles",
    "Solar Flares",
    "Solar Power",
    "Sports",
    "Standardized Testing",
    "Star Clusters",
    "Stellar Classification",
    "Stocks",
    "Strokes",
    "Sustainability",
    "Video Games",
    "Visual Arts",
    "Vollyball",
    "Wage Gap",
    "Water",
    "Water Use",
    "Weather",
    "Web Deployment",
    "Web Development",
    "Wildlife",
    "Women's Health",
    "Writing",
    "yoga"
  ],
  "tagsMap": {
    "Business and Finance": [
      "Consumer Behavior",
      "Cryptocurrency",
      "E-Commerce",
      "Economics",
      "Fraud",
      "Microfinance",
      "Options",
      "Pricing",
      "Real Estate",
      "Stocks"
    ],
    "Education": [
      "Dropout Rates",
      "Education",
      "Psych",
      "Standardized Testing"
    ],
    "Engineering": [
      "Aerospace Engineering",
      "Autonomous Vehicles",
      "Control Systems",
      "Engineering",
      "Hardware",
      "Optimization",
      "Recycling",
      "Robotics"
    ],
    "Environment": [
      "Chemicals",
      "Climate Change",
      "Earth Systems",
      "Farming",
      "Forest Fires",
      "GHGs",
      "Irrigation",
      "Natural Disasters",
      "Plants",
      "Pollution",
      "Renewable Energy",
      "Sustainability",
      "Water",
      "Water Use",
      "Weather",
      "Wildlife"
    ],
    "Healthcare and Biology": [
      "Biology",
      "Biotechnology",
      "Brain Cancer",
      "COVID-19",
      "Cancer",
      "DNA",
      "Dental Health",
      "Developmental Disorders",
      "Diabetes",
      "Diseases and Health Conditions",
      "Drugs",
      "Eyes",
      "Heart Disease",
      "Lung Cancer",
      "Mental Health",
      "Neuroscience",
      "Physiology",
      "Pneumonia",
      "Skin Cancer",
      "Strokes",
      "Women's Health"
    ],
    "Humanitarian Efforts and Social Justice": [
      "Criminal Justice",
      "Data and Privacy",
      "Drug Overdose",
      "Fake News",
      "Gender Equality",
      "Inequality",
      "Law and Order",
      "Mental Health",
      "Racial Equality",
      "Recidivism",
      "Social Justice",
      "Sociology",
      "Wage Gap"
    ],
    "Journalism": [
      "Fake News",
      "News"
    ],
    "Language and Literature": [
      "Fake News",
      "GPT-3",
      "Linguistics",
      "Modern Languages",
      "Natural Language Processing",
      "News",
      "Writing"
    ],
    "Physics": [
      "Astronomy",
      "Gamma Rays",
      "Kinematics",
      "Physics",
      "Solar Cycles",
      "Solar Flares",
      "Star Clusters"
    ],
    "PoliSci and Law": [
      "Criminal Justice",
      "Government Systems",
      "Law and Order",
      "Political Science",
      "Recidivism"
    ],
    "Psychology": [
      "Depression",
      "Facial Emotion",
      "Mental Health",
      "Psych"
    ],
    "Specific Technologies": [
      "App Development",
      "Audio Analysis",
      "BERT",
      "CNNs",
      "Classification",
      "Computer Vision",
      "Deep Learning",
      "EOL",
      "GPT-3",
      "Generative AI",
      "Google",
      "Gradient Boosting",
      "LSTMs",
      "Logistic Regression",
      "Natural Language Processing",
      "Object Detection Models",
      "Prediction",
      "Q Learning",
      "Reinforcement Learning",
      "Web Development"
    ],
    "Misc": [
      "Food and Beverage",
      "Music",
      "New York",
      "Pop Culture",
      "Product Analysis",
      "Social Media",
      "Video Games",
      "Visual Arts",
      "Yoga"
    ],
    "Published Papers and Science Fairs": [
      "Published",
      "Science Fairs"
    ],
    "Sports and Motion Sciences": [
      "Baseball",
      "Fencing",
      "Running",
      "Sports",
      "Vollyball"
    ],
    "Generative AI": [
      "LLMs",
      "Transformers"
    ]
  }
}