This repo contains a full implementation of a fully extensible multilayer perceptron. The problem is formulated as a series of matrix multiplication using numpy only.
MNIST is a standard benchmark for neural networks. With our simple implementation we get good results on this dataset.
After training our network, we use user input to classify test example. In this case the classification is live & interactive
Our multilayer perceptron uses two hidden layers with 100 units each.
Brian Dolhansky: His tutorials gave me great understanding of backpropagation algorithm Siraj Raval: The live classification was inspired from one of his tutorials