https://github.com/nshepperd/funn

Funn: Functional Neural Networks in Haskell
https://github.com/nshepperd/funn

Last synced: 3 months ago
JSON representation

Funn: Functional Neural Networks in Haskell

• Host: GitHub
• URL: https://github.com/nshepperd/funn
• Owner: nshepperd
• License: mit
• Created: 2015-07-14T14:48:14.000Z (over 9 years ago)
• Default Branch: master
• Last Pushed: 2019-04-14T19:10:15.000Z (almost 6 years ago)
• Last Synced: 2024-08-02T10:27:34.800Z (6 months ago)
• Language: Haskell
• Size: 425 KB
• Stars: 22
• Watchers: 3
• Forks: 2
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-haskell-deep-learning - funn - This is an experimental library exploring a combinator approach for building and training neural networks in haskell. | [Neil Shepperd](https://github.com/nshepperd) (Haskell Packages / Packages Under Active Development)

README

        
Funn: Functional Neural Networks in Haskell

===========================================

This is an experimental library exploring a combinator approach for building and training neural networks in haskell.

Traditional (eg. in C) libraries construct neural networks monolithically, by providing a comprehensive list of the layers' topologies to a function. The approach used in this library is an attempt at a composable system, in which networks are built by connecting smaller units together:

As `let one = fcLayer :: Network m (Blob 10) (Blob 20)` is a fully connected 10x20 layer, and `let two = sigmoidLayer :: Network m (Blob 20) (Blob 20)` is a sigmoid activation layer, we can compose them directly by feeding the output of the first into the second.

    one >>> two :: Network m (Blob 10) (Blob 20)

Parts of the interface are still quite ad-hoc and subject to change.

MHUG Talk

---------

The slides in `/mhug-talk-15` describe a mini talk I presented at the Melbourne haskell user group.