First Experimental Results of ProbCut Applied to Chess

This research explores the application of ProbCut and Multi-ProbCut (MPC) heuristics to computer chess, aiming to enhance the standard alpha-beta search algorithm. While these selective search techniques have proven highly effective in games like Othello and Shogi, their implementation in chess has historically been challenging due to the high efficacy of the existing null-move heuristic and the game's high branching factor. The study integrates these heuristics into the CRAFTY chess engine to evaluate their impact on search depth and overall playing strength.

The methodology involved fitting a linear regression model to search data from approximately 2,700 positions to establish the relationship between shallow and deep search results. Researchers then tested various depth pairs and cut thresholds to optimize performance. The implementation of MPC, which utilizes multiple depth pairs and stage-specific parameters, allowed the engine to search approximately 0.5 plies deeper than the original version of CRAFTY.

Experimental results demonstrate that MPC provides a measurable performance gain. In a 64-game match against the original CRAFTY, the MPC-enhanced version achieved a 59.4% win rate. Furthermore, tournament performance against the strong chess program YACE improved, with win rates increasing from 51% to 56% under speed chess conditions. The findings conclude that while the performance gains in chess are more modest than those observed in Othello, MPC serves as a viable and effective enhancement to existing brute-force search methods. The authors suggest that future research should focus on further fine-tuning parameters and exploring the influence of evaluation function components on search accuracy.