eulerian: This project was made in PROMYS 2018 to generate data for q-Eulerian polynomials proposed by Paul Gunnells (https://arxiv.org/pdf/1702.02446.pdf). See https://arxiv.org/pdf/1809.07398.pdf Definition 2.3, 2.6 for algorithm, also Appendix A for data. As there is no previously known efficient way to generate q-Eulerian polynomials, this program is able to generate data for n=10 with given data restraints, allowing us to make important conjectures about the properties of these polynomials.
Secondary project: CIFARtest, seer1, and resnet are a CNN, DenseNet, and ResNet, respectively, trained on CIFAR100. Final commented accuracies are ran on 10 epochs each to compare the three models.
IMPORTANT: when running these three programs, change the paths, as they save the best model for test and validation.
Densely Connected Neural Network(DenseNet) is a machine learning architecture inspired by CNN(Convolutional Neural Networks). Different from CNN, DenseNet concatenates layers of output to make sure components of inputs are connected. In doing so, intuitively speaking, DenseNets optimize their own operation layers in addition to optimizing output and thus become more and more efficient as training happens. DenseNets can deal with large data inputs without vanishing gradient and static learning rate phenomenons. We also compare the model to previous models of CNN and ResNet to test this claim.
Nathan Sun, nsun1@exeter.edu, 11th grade Not planning to go abroad. Responsibilities: proctor, Stu Lis, Symphony Orchestra
USACO Gold, CS405, 505, 590, 999. I have some experience in machine learning and deep learning (999 in class). I participate in USACO out of class and I created this program as part of my research at PROMYS to generate data for Eulerian polynomials, something which previously was unknown. See https://arxiv.org/pdf/1702.02446.pdf, https://arxiv.org/pdf/1809.07398.pdf for algorithms
I volunteered and mentored/planned HackExeter and attended USACO training sessions during the fall and winter. I really enjoyed learning how to solve USACO problems, as they are often really clever solutions. The insights I gained from USACO definitely helped me develop as a programmer. Further, I enjoy applying computer science to other areas of research- the q-Eulerian polynomial project was the best summer of my life. I would like to help organize ECC events next year and spread interest in CS, which I believe is a vastly underrated topic at Exeter. Also, I would like to organize trips to CS events outside Exeter.
Ideas for ECC next year: perhaps something like HackExeter, but for Exonians. CS is an largely overlooked field in Exeter, and perhaps more could be done to encourage people to start programming.