Research map / Multi-agent systems and games / Cooperative multi-agent RL
Multi-Agent Actor-Critic with Hierarchical Graph Attention Network
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 hierarchical graph attention representation (ablation) | HAMA · 3 vs. 3 Predator-Prey | predator score: 0.07 $\pm$0.04 -> 1.19 $\pm$0.09 | Helped |
| Add inter-agent attention (ablation) | Hierarchical graph HAMA · 3 vs. 3 Predator-Prey | predator score: 0.37 $\pm$0.06 -> 1.19 $\pm$0.09 | Helped |
| Add inter-group attention (ablation) | Hierarchical graph HAMA · 3 vs. 3 Predator-Prey | predator score: 1.03 $\pm$0.08 -> 1.19 $\pm$0.09 | Helped |
| Add attention mechanisms to the hierarchical graph (ablation) | HG-NA · 3 vs. 3 Predator-Prey | predator score: 0.57 $\pm$0.07 -> 1.19 $\pm$0.09 | Helped |
| Train HAMA predators instead of baseline predators | Predator policies against MADDPG prey · 3 vs. 3 Predator-Prey | predator score: 1.18 $\pm$0.13 -> 6.33 $\pm$0.10 | Helped |
| Transfer the policy trained with 3 agents to 50 agents | HAMA Cooperative Navigation · Cooperative Navigation transfer (n=50) | mean penalty: 0.09 -> 0.05 | Helped |
| Transfer the policy trained with 3 agents to 100 agents | HAMA Cooperative Navigation · Cooperative Navigation transfer (n=100) | mean penalty: 0.07 -> 0.04 | Helped |
| Use HAMA rather than ATOC | Cooperative Navigation · Cooperative Navigation (n=3) | mean penalty: 0.17 $\pm$0.007 -> 0.15 $\pm$0.008 | Helped |
| Use HAMA rather than MADDPG or MAAC | Predator and prey policies · 3 vs. 1 Predator-Prey | predator score per step: 0.30 $\pm$0.02 -> 0.45 $\pm$0.03; 0.35 $\pm$0.02 -> 0.39 $\pm$0.03 | Mixed |
| Use HAMA rather than MADDPG or MAAC | Predator policies · The More-The Stronger | predator score: 1.64 $\pm$0.13 -> 5.45 $\pm$0.12; 1.81 $\pm$0.29 -> 3.45 $\pm$0.30 | Helped |
Closest papers by meaning
- HGAP: Boosting Permutation Invariant and Permutation Equivariant in Multi-Agent Reinforcement Learning via Graph Attention Network (ICML 2024 · similarity 0.88)
- Towards Efficient Collaboration via Graph Modeling in Reinforcement Learning (AAAI 2025 · similarity 0.88)
- HYGMA: Hypergraph Coordination Networks with Dynamic Grouping for Multi-Agent Reinforcement Learning (ICML 2025 · similarity 0.87)
- Attention-Guided Contrastive Role Representations for Multi-agent Reinforcement Learning (ICLR 2024 · similarity 0.87)
- Inverse Attention Agents for Multi-Agent Systems (ICLR 2025 · similarity 0.87)
- Bridging Training and Execution via Dynamic Directed Graph-Based Communication in Cooperative Multi-Agent Systems (AAAI 2025 · similarity 0.87)
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.