Very interesting paper. On 9x9, life and death is critical, since white can usually win by living in two places. On 19x19, wall making moves are more important. This might explain why the patterns trained from 9x9 are so tactical and eye shape focused, and why they dont work so well on 19x19.
I didnt notice at first that the 19x19 trials were against gnugo 3.8 level 0. Why not use level 10, as in the 9x9 testing. Many Faces wins about 50% vs 3.7 level 10, and I would expect the same of Erica. Is Gnugo 3.8 so much stronger than 3.7? David -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Rémi Coulom Sent: Thursday, September 30, 2010 6:05 AM To: [email protected] Subject: [Computer-go] Monte-Carlo Simulation Balancing in Practice Hi, This is the CG'2010 paper Aja wrote with me. Abstract: Simulation balancing is a new technique to tune parameters of a playout policy for a Monte-Carlo game-playing program. So far, this algorithm had only been tested in a very artificial setting: it was limited to 5x5 and 6x6 Go, and required a stronger external program that served as a supervisor. In this paper, the effectiveness of simulation balancing is demonstrated in a more realistic setting. A state-of-the-art program, Erica, learned an improved playout policy on the 9x9 board, without requiring any external expert to provide position evaluations. Evaluations were collected by letting the program analyze positions by itself. The previous version of Erica learned pattern weights with the minorization-maximization algorithm. Thanks to simulation balancing, its playing strength was improved from a winning rate of 69% to 78% against Fuego 0.4. You can download it from there: http://remi.coulom.free.fr/CG2010-Simulation-Balancing/ Rémi _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
