Research map / Multi-agent systems and games / Cooperative multi-agent RL
Bayesian Robust Cooperative Multi-Agent Reinforcement Learning Against Unknown Adversaries
Kiarash Kazari, György Dán
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 |
|---|---|---|---|
| add learned belief about adversary type (ablation) | BATPAL · SMAC-2s3z (ACT attack) | win rate: 0.33 -> 0.72 | Improved |
| add learned belief about adversary type (ablation) | BATPAL · MPE-Spread (no attack) | normalized return: 0.81 -> 1.00 | Improved |
| replace learned belief with perfect type information (ablation) | BATPAL · SMAC-2s3z (ACT attack) | win rate: 0.34 -> 0.72 | Hurt |
| add PPO clipping to constrained adversary updates (vs plain PG) (ablation) | BATPAL · SMAC-2s3z (ACT attack) | win rate: 0.41 -> 0.72 | Improved |
| add PPO clipping to constrained adversary updates (vs plain PG) (ablation) | BATPAL · MPE-Spread (no attack) | normalized return: 0.97 -> 1.00 | No clear effect |
| use severity-partitioned adversaries instead of fixed adversary ensemble (ablation) | BATPAL · SMAC-2s3z (DYN-1 attack) | win rate: 0.14 -> 0.52 | Improved |
| use severity-partitioned adversaries instead of fixed adversary ensemble (ablation) | BATPAL · MPE-Spread (DYN-1 attack) | normalized return: 0.59 -> 0.67 | No clear effect |
| increase severity levels K from 3 to 4 (ablation) | BATPAL · SMAC-2s3z (ACT attack) | win rate: 0.16 -> 0.72 | Improved |
| increase severity levels K from 4 to 5 (ablation) | BATPAL · SMAC-2s3z (ACT attack) | win rate: 0.72 -> 0.79 | Improved |
| increase severity levels K from 4 to 5 (ablation) | BATPAL · MPE-Spread (A-EIR-MAPPO attack) | normalized return: 0.64 -> 0.73 | Mixed |
| use BATPAL instead of Gen-Maxmin against multiple victims | BATPAL · SMAC-MMM (2 victims, DYN 1) | win rate: 0.16 -> 0.44 | Improved |
| use BATPAL instead of baselines against 3 victims ACT attack | BATPAL · SMAC-MMM (3 victims, ACT) | win rate: 0.00 -> 0.00 | No clear effect |
Closest papers by meaning
- Byzantine Robust Cooperative Multi-Agent Reinforcement Learning as a Bayesian Game (ICLR 2024 · similarity 0.95)
- Wolfpack Adversarial Attack for Robust Multi-Agent Reinforcement Learning (ICML 2025 · similarity 0.90)
- Robust Multi-Agent Reinforcement Learning with Stochastic Adversary (ICML 2025 · similarity 0.90)
- Robust and Diverse Multi-Agent Learning via Rational Policy Gradient (NeurIPS 2025 · similarity 0.90)
- Empirical Study on Robustness and Resilience in Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Robust Adversarial Reinforcement Learning via Bounded Rationality Curricula (ICLR 2024 · similarity 0.89)
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.
