Home

Welcome to the Haskell_ML wiki!

Contents

• Instant Gratification

• FAQ

Description

This repository is intended to provide a robust library for exploring machine learning concepts in Haskell. In particular, it is intended to provide a gentle introduction to using Conal Elliott's ConCat library, for performing automatic differentiation in a machine learning context. The use of Conal's library in this way offers the promise of freeing us from the limitations of our current mental model for machine learning algorithms: a linear sequence of layers in which each layer consists of:

• an affine transformation with learnable parameters, followed by
• a non-linear "activation" function with unlearnable parameters.

This repository owes its initial inspiration to:

• Justin Le's blog post on building type safe, run time definable neural networks, and
• The neural package, by Lars Bruenjes, which is older and more fully developed than this repository.