A neural network implementation from scratch for CPU and GPU (in theano)
- Add preprocessing script of MNIST dataset
- Add CPU implementation (based on book by Michael Nielsen)
- Add theano implementation
- Add tests on MNIST dataset (theta.json with accuracy of 99.3% on MNIST dataset)
- Add the graphs used for choosing the final parameters (though the graphs are not labeled, it should be obvious what those graphs are telling)
A small MNIST training data is also added for testing the implementation