Implementing a Neural Network From Scratch

Based off of this video: https://www.youtube.com/watch?v=w8yWXqWQYmU

Major Modification 1:

1. Generalized functions to work for any number of layers
2. Initialized weights using Kaiming initialization Results: Accuracy after 500 epochs with layer sizes [784, 128, 64, 10] increased to 94% (from original 84%).