Category: CMUTR
Institution: Department of Computer Science, Carnegie
        Mellon University
Abstract: Golem is a simple go-playing program that seeks to play the move
        which gains the most territory. 
        In uses local tactical search to identify vulnerable stones and 
        protected liberties, and then uses this information to estimate
        the strength of each group of stones.
        Territory is computed using a relaxation procedure.
        It has a neural network trained to try to recognize 
        professional go plyers' moves based on the pattern of the 
        stones immediately surrounding each move.
        This neural network is used for move ordering and forward 
        pruning in Golem's search. 
        Golem plays at approximately the 13 kyu level.
        Its simple structure indicates a framework for future go 
        programs.
Number: CMU-CS-92-101
Bibtype: TechReport
Month: dec
Author: Herbert D. Enderton 
Title: The Golem Go Program
Year: 1992
Address: Pittsburgh, PA
Super: @CMUTR