A simple neural network written in C++ to overcome the boredom of the summer holidays.
Written to scale. Different training data may be used on the network by coding seperate shared libraries, and each network may be saved to a file (serialised) so retraining will not be necessary.
Works reasonably well on the MNIST test/training data, though a convolutional network would be necessary for more accurate number detection on foreign data.
make
make gen