Neural Network Framework

From scratch neural net implementation with customizable layers and units syntactically similar to Tensorflow. Implemented 4 different activation functions and 2 different loss functions.

Applied towards tumor classification task using Kaggle data.

DAgger Implementation

PyTorch implementation of DAgger imitation learning algorithm. Imitator trained on trajectories collected by demonstrator which was trained with the CMA-ES evolution algorithm.

Used PyRibs optimizer for CMA-ES implementation