Research map / Learning theory and algorithms / Submodular and game algorithms
Learning to Play Two-Player Perfect-Information Games without Knowledge
Quentin Cohen-Solal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use tree learning instead of terminal learning (ablation) | iterative deepening alpha-beta · Mean over 9 games | final win percentage: 29.7% -> 70.7% | Improved |
| use tree learning instead of root learning (ablation) | iterative deepening alpha-beta · Mean over 9 games | final win percentage: 38.5% -> 70.7% | Improved |
| use tree learning instead of root learning (ablation) | MCTS · Mean over 9 games | final win percentage: 61.6% -> 53.1% | Hurt |
| use tree learning instead of terminal learning (ablation) | MCTS · Mean over 9 games | final win percentage: 30.8% -> 53.1% | Improved |
| use iterative deepening alpha-beta instead of MCTS under tree learning (ablation) | tree learning · Mean over 9 games | final win percentage: 53.1% -> 70.7% | Improved |
| use Descent instead of UBFM (ablation) | tree learning · Mean over 9 games | final win percentage: 64.0% -> 82.4% | Improved |
| use Descent instead of iterative deepening alpha-beta (ablation) | tree learning · Mean over 9 games | final win percentage: 48.3% -> 82.4% | Improved |
| use Descent instead of MCTS (ablation) | tree learning · Mean over 9 games | final win percentage: 32.9% -> 82.4% | Improved |
| use Descent instead of UBFM (ablation) | tree learning · Santorini | final win percentage: 71.8% -> 69.9% | No clear effect |
| use Descent instead of UBFM (ablation) | tree learning · Surakarta | final win percentage: 69.7% -> 82.8% | Improved |
| use Descent instead of UBFM (ablation) | tree learning · Othello | final win percentage: 47.2% -> 89.4% | Improved |
| use UBFM with tree learning instead of iterative deepening alpha-beta with tree learning (ablation) | tree learning · Mean over 9 games | final win percentage: 48.3% -> 64.0% | Improved |
Closest papers by meaning
- Provable Self-Play Algorithms for Competitive Reinforcement Learning (ICML 2020 · similarity 0.87)
- Near-Optimal Reinforcement Learning with Self-Play (NeurIPS 2020 · similarity 0.87)
- Improve Agents without Retraining: Parallel Tree Search with Off-Policy Correction (NeurIPS 2021 · similarity 0.86)
- Thinking Fast and Slow with Deep Learning and Tree Search (NeurIPS 2017 · similarity 0.86)
- State-Constrained Zero-Sum Differential Games with One-Sided Information (ICML 2024 · similarity 0.84)
- From Average-Iterate to Last-Iterate Convergence in Games: A Reduction and Its Applications (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
