Date: 18/3/2009  Room: s1  Time: 14:30
Chess Revision: Acquiring the rules of Chess Variants through Theory Revision from Examples
By Aline Paes
 
Abstract:
Games are very popular as testbeds for machine learning systems, mainly because they require a focus on intelligent reasoning. The rules of a game can be meaningfully represented using the expressiveness of first-order logic. Thus, Inductive Logic Programming systems could be used to induce a classifier of such a game from a set of examples. On the other hand, quite often slightly different games are derived from a base game, originating a variant of such a game. One of the best examples of this case is the game of Chess, which has inspired the creation of different chess-like games, varying from faster, more challenging or regional versions of chess. As the knowledge acquisition is a difficult, time consuming and error prone task, if a classifier of the standard chess had been previously obtained, one could take advantage of that as starting point to obtain a classifier of a chess variant. This can be done using theory revision systems, which start with an initial theory and modify it from a set of examples, so that the final modified theory correctly represent such examples. In this talk we present a framework for obtaining a classifier of chess variants using Theory Revision from Examples. We show the improvements performed on a traditional revision system so that it would be possible to accomplish the task. Preliminary experimental results show the effectiveness of our approach.