Undergraduate research, Northeastern University, Khoury College of Computer Science, 2020
I mentored an undergraduate researcher in an independent project on semi-supervised classification. I taught him how to train neural networks in PyTorch, and we investigated image classification using label propagation in the latent space of a ResNet model. We examined the effect of different choices of distance metric/kernel function when forming the graph, and also compared the use of a simple affinity matrix against a normalized graph Laplacian matrix. We compared the performance of these models to a variety of baselines applied to a synthetic latent space constructed by dimensionality reduction on the input data.