Awesome Haskell Deep Learning

In the tradition of "awesome" (curated) lists, this is a list of references and code for doing deep learning (and adjacent/related topics) in Haskell.

• 2019 | A Functional Reboot for Deep Learning (BOB 2019 Talk) | Conal Elliott

• 2019 | Keynote: Automatic Diferentiation for Dummies (talk) | Simon Peyton Jones

• 2018 | NPFL Numerical Programming in Functional Languages (ICFP Session) 2018 Playlist | Multiple Presenters

• 2018 | Not-o-matic Differentiation | Andrew Knapp

• 2018 | Hasktorch v0.0.1 | Sam Stites

• 2018 | The Simple Essence of Automatic Differentiation (talk) | Conal Elliott

• 2018 | The Simple Essence of Automatic Differentiation | Conal Elliott

• 2018 | A Purely Functional Typed Approach to Trainable Models | Justin Le

• 2018 | Introducing the backprop library | Justin Le

• 2017 | Backprop as Functor: A compositional perspective on supervised learning | Brendan Fong, David I. Spivak, Rémy Tuyéras

• 2017 | Haskell and AI (multi-part series covering Tensorflow) | James Bowen

• 2017 | Backpack for deep learning | Kaixi Ruan

• 2017 | DeepDarkFantasy: A Programming Language for Deep Learning | Marisa Kirisame

• 2017 | Deep Learning, from a Programming Language Perspective | Marisa Kirisame

• 2016 | Computing symbolic gradient vectors with plain Haskell | Dan Aloni

• 2016 | Practical Dependent Types in Haskell (Part 2): Existential Neural Networks and Types at Runtime | Justin Le

• 2016 | Practical Dependent Types in Haskell (Part 1): Type-Safe Neural Networks | Justin Le

• 2016 | Reverse-Mode Automatic Differentiation in Haskell Using the Accelerate Library (CS240h project) | James Bradbury, Farhan Kathawala

• 2015 | Neural Networks, Types, and Functional Programming | Christopher Olah

• 2014 | Get a Brain | Ben Lynn

• 2013 | Backpropogation is Just Steepest Descent with Automatic Differentiation | Dominic Steinitz

Haskell Packages

Packages Under Active Development

• backprop - Automatic heterogeneous back-propagation that can be used either implicitly (in the style of the ad library) or using explicit graphs built in monadic style.| Justin Le
• arrayfire-haskell - High-level Haskell bindings to the ArrayFire General-purpose GPU library. | David Johnson
• backprop-hmatrix - Automatic heterogeneous back-propagation that can be used either implicitly (in the style of the ad library) or using explicit graphs built in monadic style. | Justin Le
• diffhask - DSL for forward and reverse mode automatic differentiation via a version of operator overloading. Port of DiffSharp to Haskell; currently a work in progress. | Tim Pierson
• funn - This is an experimental library exploring a combinator approach for building and training neural networks in haskell. | Neil Shepperd
• grenade - Grenade is a composable, dependently typed, practical, and fast recurrent neural network library for concise and precise specifications of complex networks in Haskell. | Huw Campbell
• gym-http-api This project provides a local REST API to the gym open-source library, includes a Haskell client by Sam Stites
• hasktorch Tensors and neural networks in Haskell, leverages the libtorch backend. | Hasktorch Contributor Team
• hnn - A neural network library implemented purely in Haskell, relying on the hmatrix library. | Alp Mestan
• rc - Reservoir computing library. | Bogdan Penkovsky
• tensor-safe - A framework to define valid deep neural network models and export them to specific languages | Leonardo Pineyro
• tensorflow - The tensorflow-haskell package provides Haskell bindings to TensorFlow. | Judah Jacobson and Greg Steuk
• TypedFlow - TypedFlow is a typed, higher-order frontend to TensorFlow and a high-level library for deep-learning. Generates python. | Jean-Philippe Bernardy

Legacy Packages

• convoluted - Dependently typed convolutional neural networks in pure Haskell. Uses the repa library for high-performance arrays, with a static wrapper that ensures networks are valid at compile-time. | Jonas Carpay
• deeplearning-hs
• dnngraph
• lambdanet
• neural - The goal of neural is to provide a modular and flexible neural network library written in native Haskell. | Lars Brünjes

Related Topics & Links

• Chris Olah's blog
• dataHaskell site and gitter chat
• idontgetoutmuch blog
• Justin Le's blog
• Monday Morning Haskell

Contribution

Feel free to send a pull request.