Last update: June 7, 2021
This repository serves the 2021 NextGen Fellows at Carnegie Mellon University under Prof. Elizabeth Holm.
We'll push new notebooks to this repository prior to our discussions (kind of like a Canvas page). Feel free to clone this repository and pull when we send word of updates.
Seminars for the first week are scheduled where we'll cover some basic concepts to bring you up to speed with some ideas and methods that we'll explore in this program.
| Day | Topic | Resources | Instructor |
|---|---|---|---|
| Monday, June 7, 2021 | Introduction: NextGen Program CV/ML for microstructure |
Overview paper Set of group papers |
Liz |
| Tuesday, June 8, 2021 | Data featurizing and dimensionality reduction by PCA | Class notes | Srujana |
| Wednesday, June 9, 2021 | Image representation using CNNs | Slides Web resources |
Bo |
| Thursday, June 10, 2021 | Clustering by k-Means and t-SNE | Class notes | Ryan |
| Friday, June 11, 2021 | Supervised machine learning with SVM | Class notes | Katelyn |
We will also go over some hands-on tutorials to help you practice new skills. These are just a start; we're excited to see where you take your research project!
| Day | Topic | Goals | Instructor |
|---|---|---|---|
| Monday, June 7, 2021 | Set up Jupyter notebooks: Acquire steel defects data and create Pandas dataframe |
-Display images by class -Data structured for subsequent analyses |
Andrew |
| Tuesday, June 8, 2021 | Traditional image representation: Represent images as brightness histograms |
-Compute and store brightness histograms for all images -Compute average brightness histogram for each class and show they are different |
Nan |
| Wednesday, June 9, 2021 | CNN image representation: Represent images using CNN layers |
-Compute and store representations of all images based on two layers of pre-trained VGG16: FC1 and shallow layers -Visualize feature vectors |
Bo |
| Thursday, June 10, 2021 | Unsupervised learning: Cluster each representation using k-means and t-SNE |
-k-means and t-SNE plots for each representation, color coded by ground truth -Compare representations and clustering methods |
Ryan |
| Friday, June 11, 2021 | Supervised learning: SVM Train and evaluate SVM |
-Classification performance (confusion matrices) for each representation | Katelyn |