Data structures and algorithms project on neural networks applied to the MNIST dataset.
The main program initializes a neural network with 300 neurons in the hidden layer and trains it in one pass through the whole training set. The error rate is calculated over all the images from the test set.