Research map / Multi-agent systems and games / Cooperative multi-agent RL
Multi-agent In-context Coordination via Decentralized Memory Retrieval
Tao Jiang, Zichuan Lin, Lihe Li, Yi-Chen Li, Cong Guan, Lei Yuan, Zongzhang Zhang, Yang Yu, Deheng Ye
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 |
|---|---|---|---|
| omit the RTG token from the embedding model (ablation) | MAICC · SMACv2: all | average return: 13.52±0.62 -> 14.51±0.46 | Improved |
| combine offline data and online buffer with exponential time decay memory (vs online only) (ablation) | MAICC · SMACv2: all | average return: 12.16±0.72 -> 14.51±0.46 | Improved |
| combine offline data and online buffer with exponential time decay memory (vs offline only) (ablation) | MAICC · SMACv2: all | average return: 11.17±0.64 -> 14.51±0.46 | Improved |
| add transition loss L_T to CEM loss (ablation) | MAICC · SMACv2: all | average return: 13.43±0.51 -> 14.51±0.46 | Improved |
| add reward loss L_R to CEM loss (ablation) | MAICC · SMACv2: all | average return: 12.32±0.48 -> 14.51±0.46 | Improved |
| add reward and transition losses to behavior loss L_mu (ablation) | MAICC · SMACv2: all | average return: 10.55±0.39 -> 14.51±0.46 | Improved |
| use hybrid utility score alpha=0.8 instead of global return only (alpha=1) (ablation) | MAICC · SMACv2: all | average return: 13.61±0.40 -> 14.51±0.46 | Improved |
| use hybrid utility score alpha=0.8 instead of predicted individual return only (alpha=0) (ablation) | MAICC · SMACv2: all | average return: 13.26±0.66 -> 14.51±0.46 | Improved |
| add centralized embedding model distilling team information into decentralized models (ablation) | MAICC-S · SMAC scenarios | average return | Improved |
| offline-only or online-only memory on grid world (ablation) | MAICC · LBF: 9x9-20s | average return | Improved |
Closest papers by meaning
- In-Context Fully Decentralized Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Coordination Connectivity: Shared Initialization Shapes the Joint-Policy Landscape in MARL (NeurIPS 2026 · similarity 0.90)
- Stateful Active Facilitator: Coordination and Environmental Heterogeneity in Cooperative Multi-Agent Reinforcement Learning (ICLR 2023 · similarity 0.89)
- Efficient Episodic Memory Utilization of Cooperative Multi-Agent Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Multi-agent cooperation through learning-aware policy gradients (ICLR 2025 · similarity 0.89)
- Learning to Share: Selective Memory for Efficient Parallel Agentic Systems (ICML 2026 · 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.
