neural_network

This is a school assignment completed in the Fall semester of 2020. It is a ground up implementation of a neural network designed to train on the fashion MNIST dataset. The implementation was finished in Rust and contains a variety of optimization features, such as ADAM, dropout, learning rate decay, L2 regularization and parallelized matrix multiplication. The minimal required accuracy was 88% and maximum training time 30 minutes on the school server. Current implementation can train small networks to ~88% in 5 minutes.