Research map / Multi-agent systems and games / Cooperative multi-agent RL
Adversarial Attacks on Multi-Agent Communication
James Tu, Tsunhsuan Wang, Jingkang Wang, Sivabalan Manivasagam, Mengye Ren, Raquel Urtasun
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 |
|---|---|---|---|
| adversarially train against the white box feature-level attacker | multi-agent detector · ShapeNet (perturbed) | AP@0.7: 0.62 -> 66.00 | Improved |
| adversarially train against the white box feature-level attacker | V2VNet · V2V (perturbed) | AP@0.7: 7.55 -> 83.44 | Improved |
| adversarially train against the white box attacker (clean data) | V2VNet · V2V (clean) | AP@0.7: 82.19 -> 82.60 | Improved |
| train surrogate with domain adaptation (DA) for transfer attack (ablation) | transfer attack · ShapeNet (2 agents) | victim AP@0.7 (lower is stronger attack): 66.21 -> 42.59 | Improved |
| add ILAP projection on top of DA (ablation) | transfer attack · ShapeNet (2 agents) | victim AP@0.7: 42.59 -> 35.69 | Improved |
| add diversified inputs (DI) on top of DA (ablation) | transfer attack · ShapeNet (2 agents) | victim AP@0.7: 42.59 -> 49.38 | Hurt |
| reuse the previous perturbation across frames (ablation) | online attack · V2V (2 agents) | victim AP@0.7: 56.98 -> 7.55 | Improved |
| warp the perturbation with egomotion rigid transform (ablation) | online attack · V2V (6 agents) | victim AP@0.7: 77.37 -> 76.18 | Improved |
| use adversarial loss L_adv instead of negative task loss (ablation) | white box attack · ShapeNet (2 agents) | victim AP@0.7: 6.10 -> 0.37 | Improved |
| use adversarial loss L_adv instead of negative task loss (ablation) | white box attack · V2V (2 agents) | victim AP@0.7: 20.8 -> 7.55 | Improved |
| use GNN fusion instead of mean pooling for robustness (ablation) | V2VNet · V2V (4 agents, perturbed) | AP@0.7: 12.93 -> 52.31 | Improved |
| use median instead of mean pooling in GNN nodes (ablation) | V2VNet · V2V (6 agents) | AP@0.7 clean vs perturbed: 92.94 -> 90.75 clean; 76.18 -> 86.30 perturbed | Mixed |
Closest papers by meaning
- Dissecting Adversarial Robustness of Multimodal LM Agents (ICLR 2025 · similarity 0.87)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.86)
- AdvEvo-MARL: Shaping Internalized Safety through Adversarial Co-Evolution in Multi-Agent Reinforcement Learning (ICML 2026 · similarity 0.85)
- When Embedding-Based Defenses Fail: Rethinking Safety in LLM-Based Multi-Agent Systems (ICML 2026 · similarity 0.85)
- MaMa: A Game-Theoretic Approach for Designing Safe Agentic Systems (ICML 2026 · similarity 0.85)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · 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.
