KrwEmd: Revising the Imperfect Recall Abstraction from Forgetting Everything
Keywords: game theory, imperfect-information games, games with ordered signals, computer poker, imperfect-recall abstraction, unsupervised learning
TL;DR: The paper introduces KrwEmd, an algorithm addressing excessive abstraction in games like Texas Hold'em, caused by extreme imperfect recall, which harms AI performance. KrwEmd effectively enhances AI gameplay compared to previous methods.
Abstract: Excessive abstraction is a serious issue in solving games with ordered signals—a subset of imperfect information games, caused by extreme implementations of imperfect recall, which discard all historical information and, as a result, negatively impact AI performance. This paper presents KrwEmd, the first practical algorithm designed to address this issue. We first introduce the k-recall winrate feature, which not only qualitatively distinguishes signal infosets by leveraging future and, more importantly, historical game information, but also quantitatively reflects their similarity. We then build on this by developing the KrwEmd algorithm, which cluster signal infosets using Earth Mover’s Distance to assess discrepancies between their features. Experimental results demonstrate that KrwEmd significantly enhances AI gameplay performance compared to existing algorithms.
Supplementary Material: zip
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5250
Loading