A simple neural network to recognize handwritten digits. This project is based off this tutorial by Michael Nielson and this Youtube series on neural networks. This project was meant as a way to learn about neural networks, so the only reference used was the theory and equations described.
The network was trained on the mnist handwritten digits database.
The trained network currently has an accuracy of around 94% on the training data, and 93% on the test data.
Currently the network is fairly basic. Michael Nielson's tutorial goes over several ways to improve the network, including a new cost function.