Research map / Learning theory and algorithms / Submodular and game algorithms
Safe Opponent-Exploitation Subgame Refinement
Mingyang Liu, Chengjie Wu, Qihan Liu, Yansen Jing, Jun Yang, Pingzhong Tang, Chongjie Zhang
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 SES reach-distribution exploitation instead of EXP-STRATEGY fixed opponent strategy, alpha=0.7 (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | exploitability: 134.37 -> 41.95 | Improved |
| use SES instead of EXP-STRATEGY, alpha=1.0 (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | exploitability: 981.47 -> 55.10 | Improved |
| use SES instead of EXP-STRATEGY, alpha=0.3 (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | exploitability: 34.75 -> 28.06 | Improved |
| use SES instead of EXP-STRATEGY, alpha=0.7 (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | evaluation payoff: -27.22 -> 6.00 | Improved |
| use SES instead of EXP-STRATEGY, alpha=1.0 (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | evaluation payoff: -88.94 -> 3.30 | Improved |
| use SES instead of EXP-STRATEGY, alpha=0.3 (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | evaluation payoff: 1.63 -> 6.21 | Improved |
| use SES instead of EXP-STRATEGY at alpha=0.0 (pure safety) (ablation) | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | exploitability and evaluation: 29.43 -> 29.43 | No clear effect |
| increase exploitation level alpha from 0.0 to 1.0 in SES | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | exploitability: 29.43 -> 55.10 | Hurt |
| increase exploitation level alpha from 0.0 to 0.3 in SES | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | evaluation payoff: 5.15 -> 6.21 | Improved |
| increase alpha from 0.3 to 1.0 in SES under estimation error 0.1 | SES · FHP, Pr_shuffle=0.3, estimation error 0.1 | evaluation payoff: 6.21 -> 3.30 | Hurt |
| increase estimation error epsilon of opponent model (0.0 to 1.2) | SES · Leduc and FHP, Pr_shuffle 0.0/0.3/0.7 | exploitability and payoff | Hurt |
| increase alpha with accurate opponent estimate (epsilon=0.0) | SES · FHP head-to-head | expected payoff | Improved |
Closest papers by meaning
- Safe and Robust Subgame Exploitation in Imperfect Information Games (ICML 2024 · similarity 0.92)
- The Update-Equivalence Framework for Decision-Time Planning (ICLR 2024 · similarity 0.87)
- Advantage Alignment Algorithms (ICLR 2025 · similarity 0.85)
- Offline Opponent Modeling with Truncated Q-driven Instant Policy Refinement (ICML 2025 · similarity 0.85)
- The Attacker in the Mirror: Breaking Self-Consistency in Safety via Anchored Bipolicy Self-Play (NeurIPS 2026 · similarity 0.85)
- Designing Skill-Compatible AI: Methodologies and Frameworks in Chess (ICLR 2024 · similarity 0.85)
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.
