Author: Michael Norrish
Source: GZipped PostScript (102kb); Adobe PDF (513kb)
This report describes a new method for knowledge-intensive learning in the domain of two-person turn-taking complete information games. The method acquires general patterns from examples of incorrect play that can be used to avoid similar mistakes in the future. The results of this research are embodied in ELGAR, a system that acquires patterns for the games of noughts and crosses and go-moku.