Research map / Multi-agent systems and games / Cooperative multi-agent RL
Offline Multi-Agent Reinforcement Learning via Sequential Score Decomposition
Dan Qiao, Wenhao Li, Shanchao Yang, Hongyuan Zha, Baoxiang Wang
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 |
|---|---|---|---|
| replace independent per-agent behavior regularization with sequential score decomposition (ablation) | behavior-regularized policy optimization · Bandit toy (2-Gaussian mixture) | evaluation reward: 0±1 -> 1±0 | Improved |
| use sequential conditional scores rather than centralized critic with marginal regularization (ablation) | BRPO-CTDE · Bandit toy (2-Gaussian mixture) | evaluation reward: 0±1 -> 1±0 | Improved |
| sequential score decomposition versus joint action learning (ablation) | OMSD · Bandit toy (2-Gaussian mixture) | evaluation reward: 1±0 vs 1±0 | No clear effect |
| OMSD versus best baselines on MPE, averaged over 9 datasets | offline MARL on MPE · MPE (9 datasets, average) | average normalized score: 87.3 -> 126.7 | Improved |
| OMSD on lower-quality (random, medium) MPE data | offline MARL on MPE · MPE Predator Prey Random | normalized score: 78.5 -> 133.9 | Improved |
| OMSD on expert Cooperative Navigation data | offline MARL on MPE · MPE Cooperative Navigation Expert | normalized score: 126.3 -> 102.3 | Hurt |
| OMSD on MaMuJoCo 2-Ant and 4-Ant Good datasets | offline MARL on MaMuJoCo · MaMuJoCo 2-Ant Good | episode return: 2946 -> 2714 | Hurt |
| OMSD on OMIGA MaMuJoCo benchmark | offline MARL on MaMuJoCo · MaMuJoCo OMIGA datasets (average) | average episode return: 1954.74 -> 3400 | Improved |
| add OMSD post-training to pretrained IQL critic/policy | pretrained IQL · MaMuJoCo (average across datasets) | episode return | Improved |
| randomize agent update order in sequential decomposition (ablation) | OMSD · OMIGA Hopper-v2 | episode return | No clear effect |
| vary regularization coefficient beta over 0.001 to 0.5 (ablation) | OMSD · MaMuJoCo / MPE | episode return | Mixed |
Closest papers by meaning
- ComaDICE: Offline Cooperative Multi-Agent Reinforcement Learning with Stationary Distribution Shift Regularization (ICLR 2025 · similarity 0.92)
- Offline Multi-Agent Reinforcement Learning via In-Sample Sequential Policy Optimization (AAAI 2025 · similarity 0.91)
- Cooperative Policy Agreement: Learning Diverse Policy for Offline MARL (AAAI 2025 · similarity 0.90)
- Dispelling the Mirage of Progress in Offline MARL through Standardised Baselines and Evaluation (NeurIPS 2024 · similarity 0.89)
- Exploiting Structure in Offline Multi-Agent RL: The Benefits of Low Interaction Rank (ICLR 2025 · similarity 0.88)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · 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.
