This is a Neural Network I created from scratch to get a better understanding of the concept. I documented everything, in an easy to follow document with bite sized chunks of the concepts that were required to build this network.
The dataset used was the MNIST handwritten digits dataset, and the network has the below illustrated architecture.
I trained the network using 2 different optimizers - Gradient Descent defined in gradient_descent
function and Gradient Descent with Momentum defined in gradient_descent_wm
function
to study whether the convergence of the network.