Research map / Multi-agent systems and games / Cooperative multi-agent RL
Boosting Multi-agent Reinforcement Learning via Contextual Prompting
Yue Deng, Zirui Wang, Xi Chen, Yin Zhang
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 |
|---|---|---|---|
| add contextual prompting (tree) to QMIX | QMIX · Particle World Tag, 700k steps | cumulative return: 56.05 ± 8.03 -> 153.83 ± 20.12 | Improved |
| add contextual prompting (tree) to IQL | IQL · Particle World Tag, 700k steps | cumulative return: 31.73 ± 10.86 -> 95.38 ± 18.37 | Improved |
| add contextual prompting (tree) to QMIX | QMIX · Particle World Spread, 5k steps | cumulative return: −160.66 ± 11.22 -> -156.18 ± 7.56 | Improved |
| add contextual prompting (tree) to QMIX | QMIX · Particle World Reference, 5k steps | cumulative return: −84.38 ± 9.92 -> -57.81 ± 5.73 | Improved |
| add contextual prompting (tree) to IQL | IQL · Particle World Spread, 5k steps | cumulative return: −214.48 ± 6.36 -> −173.11 ± 19.56 | Improved |
| add contextual prompting (tree) to QMIX | QMIX · Particle World, average | cumulative return: −57.78 ± 11.71 -> −29.19 ± 10.59 | Improved |
| add contextual prompting (tree) to IQL | IQL · Particle World, average | cumulative return: −78.53 ± 20.96 -> −41.61 ± 7.36 | Improved |
| use MCTS trajectory tree instead of replay-buffer trajectory pool for training the predictor (ablation) | prompting QMIX · Particle World, average | cumulative return: −60.97 ± 11.81 -> −29.19 ± 10.59 | Improved |
| use trajectory tree instead of pool (ablation) | prompting IQL · Particle World Tag, 700k steps | cumulative return: 28.71 ± 10.18 -> 95.38 ± 18.37 | Improved |
| use trajectory pool prompting instead of plain IQL (ablation) | IQL · Particle World Reference, 700k steps | cumulative return: −50.81 ± 55.75 -> −38.45 ± 3.18 | Mixed |
| replace GRU contextual predictor with transformer predictor (ablation) | prompting QMIX · Spread, Reference, Tag | convergence speed: about twice slower | Hurt |
| replace GRU predictor with transformer predictor (ablation) | prompting QMIX · Tag | return | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- QMIX — QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (ICML 2018) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Efficient Multi-agent Reinforcement Learning by Planning (ICLR 2024 · similarity 0.88)
- Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective (NeurIPS 2025 · similarity 0.88)
- Foundations of Multi-Agent Learning in Dynamic Environments: Where Reinforcement Learning Meets Strategic Decision-Making (AAAI 2025 · similarity 0.88)
- Dispelling the Mirage of Progress in Offline MARL through Standardised Baselines and Evaluation (NeurIPS 2024 · similarity 0.88)
- Revisiting Cooperative Off-Policy Multi-Agent Reinforcement Learning (ICML 2025 · similarity 0.88)
- Adaptive Context Length Optimization with Low-Frequency Truncation for Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
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.
