Research map / Adversarial robustness and security / Black-box and transfer attacks
Adversarial Attack on Black-Box Multi-Agent by Adaptive Perturbation
Jianming Chen, Yawen Wang, Junjie Wang, Xiaofei Xie, Yuanzhe Hu, Qing Wang, Fanjiang Xu
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 |
|---|---|---|---|
| adaptive victim and malicious-action selection with single perturbed agent vs MASafe perturbing all agents | AdapAM · SMAC-1c3s5z (normal) | reward under attack: 9.36 -> 9.02 (MASafe -> AdapAM, lower is better) | Improved |
| attack with AdapAM vs MASafe in large-agent SMAC maps | AdapAM · SMAC-bane vs bane (normal) | reward under attack: 12.25 (MASafe) vs 13.21 (AdapAM) | Hurt |
| attack with AdapAM vs MASafe | AdapAM · SMAC-27m vs 30m (normal) | win rate under attack: 23% (MASafe) vs 28% (AdapAM) | Hurt |
| AdapAM on robust (ROMANCE) MAS vs baselines | AdapAM · SMAC-1c3s5z (robust) | reward under attack: 10.79 -> 9.12 (MASafe -> AdapAM) | Improved |
| AdapAM on robust MAS in bane vs bane | AdapAM · SMAC-bane vs bane (robust) | reward under attack: 13.05 (MASafe) vs 13.73 (AdapAM) | Mixed |
| perturb only one adaptively chosen agent | AdapAM · SMAC-8m | L-inf perturbation magnitude: 0.33 -> 0.12 (MASafe -> AdapAM) | Improved |
| perturb only one adaptively chosen agent | AdapAM · SMAC-27m vs 30m | detection F1: 0.91 -> 0.36 (MASafe -> AdapAM) | Improved |
| select victim and action adaptively vs fixed victim (Lin) or all agents | AdapAM · MPE-spread (normal) | reward under attack: -968.30 (Lin) -> -1059.49 | Improved |
Closest papers by meaning
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.86)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.86)
- When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems (ICML 2026 · similarity 0.85)
- AdvEvo-MARL: Shaping Internalized Safety through Adversarial Co-Evolution in Multi-Agent Reinforcement Learning (ICML 2026 · similarity 0.85)
- Robust and Diverse Multi-Agent Learning via Rational Policy Gradient (NeurIPS 2025 · similarity 0.85)
- Hierarchical Attacks for Multi-Modal Multi-Agent Reasoning (CVPR 2026 · 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.
