The abstract is read at the publisher; the summary is JClub's.
Ataraxos, an AI for Stratego, defeated the most decorated human Stratego player by a large margin, achieving a first superhuman result while using orders of magnitude less compute and data.
- Why it matters: Hidden information can overwhelm established reinforcement learning and search, leaving top-level play in large-scale imperfect-information games like Stratego beyond AI reach.
- What they did: Ataraxos combines self-play reinforcement learning with test-time search under hidden information, then applies the same techniques to build superhuman Barrage Stratego and strong Hanabi and dou dizhu agents.
- The result: The approach yields superhuman performance across adversarial, cooperative, and team games, suggesting a scalable design pattern for reinforcement learning and search under large hidden information.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.