Neuron
A Haskell MNIST recognition neural network trainer and solver.
I implemented this a few years ago when I was trying to learn the basics of neural networks and machine learning. It's a feed-forward neural network, optimised using gradient descent. I also implemented an accelerated version using matrices and hmatrix.
It's quite pleasant to implement this kind of thing in haskell since a lot of the required calculations are implemented naturally as pure functions, and in some cases it's possible to use automatic differentation to derive the necessary derivatives.
Viewing project details for Neuron, take me back to projects