Research map / Multi-agent systems and games / Cooperative multi-agent RL
Reward Poisoning Attacks on Offline Multi-Agent Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a joint multi-agent reward-poisoning attack that makes the target policy an MPDSE instead of attacking each agent separately | Offline MARL · Full-coverage dataset, target policy (1, 1) | L1 attack cost: 0 vs nonzero | Helped |
| Poison rewards to make the target policy an iota-strict MPDSE across plausible games | Offline MARL · Plausible Markov Games induced by the poisoned dataset | Target-policy adoption guarantee | Helped |
| Require full coverage of every state, joint action, and period | Offline MARL reward-poisoning attack · State-action-period tuples | Attack feasibility | Helped |
| Increase dataset visitation counts so reward confidence intervals are narrower | Offline MARL reward-poisoning attack · State-action-period tuples | Attack feasibility | Helped |
| Account for reward and transition uncertainty with Q confidence-bound backward induction | Offline MARL reward-poisoning attack · Finite-horizon Markov Games | Target-policy guarantee | Helped |
| Reduce the attack to separate single-agent poisoning problems | Offline MARL reward-poisoning attack · Full-coverage dataset, target policy (1, 1) | L1 attack cost: 0 vs nonzero | Hurt |
| Increase transition uncertainty enough that plausible transition sets include uniform transitions | Offline MARL reward-poisoning attack · Markov Games with uniform transitions in each period's confidence set | L1 attack cost | Hurt |
Closest papers by meaning
- Data Poisoning to Fake a Nash Equilibria for Markov Games (AAAI 2024 · similarity 0.92)
- Optimal Attack and Defense for Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank (ICLR 2025 · similarity 0.88)
- Multi-level Certified Defense Against Poisoning Attacks in Offline Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Offline Multi-Agent Reinforcement Learning via In-Sample Sequential Policy Optimization (AAAI 2025 · similarity 0.87)
- Dispelling the Mirage of Progress in Offline MARL through Standardised Baselines and Evaluation (NeurIPS 2024 · 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.