AI

Introduction

This is the home for the Haskell AI Strike Force! Here we will collect code, problems, papers, ideas, and people for putting together a flexible AI toolkit in Haskell.

People

If interested in contributing to or monitoring this project, please put your name, nickname (if applicable - e.g., if you talk on #haskell), and email address so we can keep each other up-to-date.

Andrew Wagner (chessguy) <wagner dot andrew at gmail>

Bryan Green (shevek) <dbryan dot green at gmail>

Ricardo Herrmann <rherrmann at gmail>

Dan Doel (dolio) <dan dot doel at gmail>

Chung-chieh Shan (ccshan) <ccshan at cs dot rutgers dot edu>

Adam Wyner (Lawman) <adam dot wyner dot info>

Allan Erskine (thedatabase) <allan dot erskine at gmail>

Dave Tapley (DukeDave) <dukedave at gmail>

Lloyd Allison <lloyd dot allison at infotech dot monash dot edu dot au>

Paul Berg (Procyon) <procyon at procyondevelopments dot com>

Eric Kow (kowey) <eric dot kow at gmail> [watching on the sidelines]

Charles Blundell <blundellc at gmail>

Mathew Mills (mathewm) <mathewmills (at) gmail (dot) com>

Jason Morton (inverselimit) <jason.morton at gmail>

Jiri Hysek (dvekravy) <xhysek02 at stud dot fit dot vutbr dot cz> [NN, EA]

Shahbaz Chaudhary <shahbazc at gmail> [interested in GP]

Hans van Thiel <hthiel dot char á zonnet tot nl> [automated rule discovery, author of the Emping data mining utility]

Alp Mestanogullari (Alpounet) <alp (at) mestan (dot) fr> [machine learning mainly]

Chris Pettitt (cpettitt) <cpettitt at gmail>

Nathaniel Neitzke (nneitzke) <nightski at gmail>

Ideas

• In short, parts of this project can range from established ideas to new syntheses. ccshan: The high level of domain-specific abstraction that Haskell enables is ideal for AI, because AI programs are often "meta": we need to model agents who model the world, and sometimes to model agents who model agents who model the world, etc. In particular, monads are a good way to structure and solve decision processes, as I've started to explore as part of a course on computational modeling that I'm teaching. Given that Haskell is a good language for modular interpreters and compilers, it would also be nice to create and refactor in Haskell an implementation of a rational programming language like Avi Pfeffer's IBAL -- not only is probability distribution a monad, I just realized that a certain kind of variable elimination is simply garbage collection in a call-by-need language!

Things that need a home

If there are things that should be included in the project, but you're not sure where it should go, place it here! I'll start with:

• http://catenova.org/~awagner/Simplifier
  • This was given to me by Alfonso Acosta (mentioned recently on haskell-cafe)

• http://catenova.org/~awagner/GPLib
  • GPLib is a work in progress by yours truly, hopefully a future framework for genetic algorithms in haskell.

• http://www.haskell.org/haskellwiki/Libraries_and_tools/Linguistics

I've proposed a machine learning library for this year's Google Summer of Code. [1] There has been a few interested (and seemingly well qualified) students, too. I'm not sure if it qualifes as "AI", but if you are interested in this project (as a potential student, mentor, or just...well, interested), please add yourself to the above link, and/or get in touch with me at <ketil at malde dot org>. --Ketil 07:46, 26 March 2007 (UTC)

Martin Erwig's probabilistic functional programming (PFP) project, including an implementation of the probability monad:

• http://web.engr.oregonstate.edu/~erwig/pfp/

Culmination of some recent posts about the probability monad on Random Hacks (including a darcs repository):

• http://www.randomhacks.net/articles/2007/03/03/smart-classification-with-haskell

sigfpe's coverage and highly algebraic view of the probability monad in Haskell:

• http://sigfpe.blogspot.com/2007/02/monads-for-vector-spaces-probability.html

Two links I found today that are interesting:

• http://perception.inf.um.es/darcs/darcsweb.cgi
• http://www-student.cs.york.ac.uk/~cb224/

Polytypic unification - unification seems particularly useful for AI tasks (at least natural language stuff)... wouldn't be nice to have a generic library that does it for you?

• http://www.cs.chalmers.se/~patrikj/poly/unify/

Easy-to-use work-in-progress neural network library, by Alp Mestan and Chaddaï Fouché :

• http://github.com/alpmestan/HNN/tree/master

Proposed Module Hierarchy

• AI
  • AI.Searching
    • AI.Searching.Evolutionary
  • AI.Logic
  • AI.Planning
  • AI.Probabilistic
  • AI.Learning
    • AI.Learning.Kernel
    • AI.Learning.NeuralNet
  • AI.Classification
    • AI.Classification.ExpertSystem
  • AI.Communication

Proposed sample format for a wiki page on a topic or sub-topic

AI/Logic/Fuzzy

The slashes show that Logic is a subpage of AI and Fuzzy is a subpage of AI/Logic. MediaWiki will then generate links back up the chain of pages. (Try the links to see)

• Fuzzy logic is blah blah...
• Sub-topics:
  • Trivial fuzzy logic in Haskell
  • Type 2 fuzzy logic
• Links to existing literature:
  • General
    • My first fuzzy logic book
  • Specific to functional programming / Haskell
    • Fun with fuzzy functions
• Typical problems:
  • Problem 1: blah blah blah
  • Problem 2: blah blah blah
• List of people involved in the area
  • Me
  • Someone else
• Body
  • List of goals
  • Progress being made on them
  • Code and documentation.

Current sub-pages

• /Logic/Fuzzy
• /Genetic programming/Evolutionary chess
• /Genetic programming/GPLib