MAPLE: Multi-State Aggregated Policy Evaluation for AlphaZero in Imperfect-Information Games
The paper introduces MAPLE, a new method for evaluating policies in imperfect-information games using a tree search approach. This method combines the strengths of existing techniques while controlling computational costs. Experiments demonstrate that MAPLE significantly improves performance over the traditional AlphaZero baseline in specific games.
- ▪MAPLE stands for Multi-State Aggregated Policy Evaluation.
- ▪It aggregates policy and value evaluations from multiple sampled world states within a single search tree.
- ▪Experiments on Phantom Go and Dark Hex show Elo improvements of 291 and 136, respectively.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.24139 (cs) [Submitted on 22 May 2026] Title:MAPLE: Multi-State Aggregated Policy Evaluation for AlphaZero in Imperfect-Information Games Authors:Qian-Rong Li, Hung Guei, I-Chen Wu, Ti-Rong Wu View a PDF of the paper titled MAPLE: Multi-State Aggregated Policy Evaluation for AlphaZero in Imperfect-Information Games, by Qian-Rong Li and 3 other authors View PDF Abstract:Imperfect-information games (IIGs) are challenging, as players must make decisions without fully observing the true game state. While AlphaZero has achieved remarkable success in perfect-information games, extending it to IIGs remains difficult.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
