Dynamic Randomization Enhances Monte-Carlo Go

This paper proposes two dynamic randomization techniques for Monte-Carlo Go that uses Monte-Carlo tree search with UCT algorithm. First, during the in-tree phase of a simulation game, the parameters are randomized in selected ranges before each simulation move. Second, during the playout phase, the order of simulation move generators are hierarchically randomized before each playout move. Both dynamic randomization techniques increase diversity while keeping the sanity of the simulation game. The first technique, dynamic randomization of the parameters, increase the winning percentage of the author´s program GoIntellect(GI) against GnuGo3.8 by 8 percentage points on the average in 19×19 games. The second technique used in conjunction with the first technique further increases the winning percentage of GI against GnuGo3.8 by up to an additional 7+ percentage points in 19×19 games.