Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Off-Belief Learning
Hengyuan Hu, Adam Lerer, Brandon Cui, Luis Pineda, Noam Brown, Jakob Foerster
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 |
|---|---|---|---|
| Apply OBL with a uniform-random starting policy to learn a grounded policy | Hanabi agents · 2-player Hanabi | Cross-Play score: 20.85 $\pm$ 0.03 vs 21.77 $\pm$ 0.68 | Improved |
| Iterate OBL one level, using level 1 to generate beliefs (ablation) | OBL · 2-player Hanabi | Cross-Play score: 20.85 $\pm$ 0.03 -> 23.24 $\pm$ 0.03 | Improved |
| Iterate OBL from level 2 to level 3 (ablation) | OBL · 2-player Hanabi | Cross-Play score: 23.24 $\pm$ 0.03 -> 23.68 $\pm$ 0.05 | Improved |
| Iterate OBL from level 3 to level 4 (ablation) | OBL · 2-player Hanabi | Cross-Play score: 23.68 $\pm$ 0.05 -> 23.76 $\pm$ 0.06 | Improved |
| Iterate OBL from level 3 to level 4 (ablation) | OBL · 2-player Hanabi | Score with Clone Bot: 16.54 $\pm$ 0.28 -> 16.76 $\pm$ 0.16 | Improved |
| Iterate OBL from level 2 to level 3 (ablation) | OBL · 3-player Hanabi | Score with Clone Bot: 14.09 $\pm$ 0.48 -> 13.96 $\pm$ 0.47 | Hurt |
| Iterate OBL from level 3 to level 4 (ablation) | OBL · 3-player Hanabi | Score with Clone Bot: 13.96 $\pm$ 0.47 -> 13.88 $\pm$ 0.47 | Hurt |
| Use OBL rather than Other-Play | Hanabi agents · 3-player Hanabi | Cross-Play score: 17.36 $\pm$ 0.19 -> 23.02 $\pm$ 0.01 | Improved |
| Use OBL rather than Other-Play | Hanabi agents · 3-player Hanabi | Score with Clone Bot: 12.18 $\pm$ 0.25 -> 14.09 $\pm$ 0.48 | Improved |
| Add an auxiliary playable/discardable/unknown prediction task (ablation) | Hanabi self-play baselines · 2-player Hanabi | Self-Play score: 24.27 $\pm$ 0.01 vs 24.26 $\pm$ 0.01 | No clear effect |
| Add color-shuffling data augmentation (ablation) | Clone Bot · 2-player Hanabi | Self-Play score: 19.93 $\pm$ 0.09 -> 21.01 $\pm$ 0.07 | Improved |
Closest papers by meaning
- Model-based Reinforcement Learning for Confounded POMDPs (ICML 2024 · similarity 0.87)
- ODRL: A Benchmark for Off-Dynamics Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.87)
- Independent Learning of Nash Equilibria in Partially Observable Markov Potential Games with Decoupled Dynamics (NeurIPS 2026 · similarity 0.86)
- Statistical Tractability of Off-policy Evaluation of History-dependent Policies in POMDPs (ICLR 2025 · similarity 0.86)
- Offline Multi-Agent Reinforcement Learning via In-Sample Sequential Policy Optimization (AAAI 2025 · similarity 0.86)
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.
