Advertisement

Training a neural net as an evaluation fonction ?

Started by March 23, 2010 01:26 PM
2 comments, last by Predictor 14 years, 7 months ago
Hi everyone. I am looking for some background articles on the use of neural networks to provide an evaluation function in a two-players board game. The main reason is that I recently found this link : http://keldon.net/bluemoon/ where you can you can find a very good ai for a card game called bluemoon. On its turn, the ai tries every possible move and evaluates the resulting situations with a neural network. What surprised me is the way the networks are trainned, with only minimal prior knowledge on the strategy, by having two networks play a lot of games against each other. (and I was also impressed by the efficiency of this approach : the ia is not an easy opponent !) So do you know any other similar works ? Is this kind of training commonly used ? ps : please forgive my bad english.
There is a long history of using ANNs as evaluation functions for backgammon. The Wikipedia page about the game is probably a good place to start.

Besides that, I know of an attempt by David Fogel at doing this for checkers, although the experiment wasn't setup cleanly enough to satisfy my standards for calling it a success. For instance, they never compared the evaluation function they obtained with evaluation functions obtained by any other method. For all I know, their success in playing against amateurs online just means that depth-6 search with material+noise as evaluation function works fine against those opponents.

Oh, and your English is perfectly understandable.
Advertisement
Thank you for your answer !
The training scheme looks indeed the same as the one used for backgammon and described there : http://www.research.ibm.com/massive/tdl.html
Quote: Original post by Spacepeon
I am looking for some background articles on the use of neural networks to provide an evaluation function in a two-players board game.


The main reason is that I recently found this link : http://keldon.net/bluemoon/
where you can you can find a very good ai for a card game called bluemoon. On its turn, the ai tries every possible move and evaluates the resulting situations with a neural network.
What surprised me is the way the networks are trainned, with only minimal prior knowledge on the strategy, by having two networks play a lot of games against each other.
(and I was also impressed by the efficiency of this approach : the ia is not an easy opponent !)

So do you know any other similar works ? Is this kind of training commonly used ?


There have been a number of efforts along these lines. Look for systems like BOXES and Adaptive Critic.

This topic is closed to new replies.

Advertisement