California State University San Bernardino
Department of Computer Science and Engineering
Masters Thesis Defense

Date 05/22/2009 Time 2:00pm-3:00pm Location JBH 389/391

Title FABLE: Finite Automata Based Learning Engine

Candidate Ryan Jackson 
Advisor Dr. George M. Georgiou 
Committee Members Dr. Kerstin Voigt Dr. Ernesto Gomez 


Abstract

FABLE is an attempt to create a system that learns how to solve problems 
without prior knowledge.  It does this by building FA models of the 
problem.  It uses the FA models to make decisions that are more likely to 
lead to a desired goal state.  The results of each choice are recorded and
used to improve the FA model for future use.  Current research into the 
automated building of FA models has been focused on recognizing the same 
strings as a target FA with an accuracy that is PAC.  FABLE attempts to use
automated building of FA models for the purpose of improving outcomes for 
specific problems, whether the FA built is PAC or not.  FABLE is tested on
the problems of minesweeper, tic-tac-toe, and checkers.  The results show 
that in all three cases, FABLE learned to play significantly better than 
the results expected from random play within less than 100 games.