This is an implementation of a feed-forward backpropogating neural network in Matlab. It achieves over 95% recognition accuracy on MNIST digit recognition data.

To train it, use the following code:

s = rng; % Store random number generator state
params = test; % Train network on random subset (and do cross-validation)
params.iterations = 300; % Set iterations
rng(s);
theta = test(params); % Train final network
rng(s);
test(theta); % Test final network