Implementation of basic Neural Network in Numpy

The implementation is similar to Keras implementation of Neural Network. Layers include Batch Norm, Convolutional 2D, Maxpool 2D, etc.

Notes:

1. Does not take advantage of GPU processing
2. Not optimized for running on large datasets
3. LSTM and GRU implementations are future work
4. train.py and validate.py were run on MNIST dataset