This repo contains the code for a project I worked on to investigate how deformable convolution could help make MNIST classifiers be more robust to geometric perturbations in the input digits.
The project was for the Fall 2019 course ECBM E4040: Neural Networks & Deep Learning at Columbia University, taught by Prof. Zoran Kostic.