Simple NN from scratch trained on MNIST dataset

Our NN will have a simple two-layer architecture. Input layer a[0] will have 784 units corresponding to the 784 pixels in each 28x28 input image. A hidden layer a[1] will have 10 units with ReLU activation, and finally our output layer a[2] will have 10 units corresponding to the ten digit classes with softmax activation. Inspired by Samson Zhang.