Research map / Multi-agent systems and games / Cooperative multi-agent RL
SPD: Synergy Pattern Diversifying Oriented Unsupervised Multi-agent Reinforcement Learning
Yuhang Jiang, Jianzhun Shao, Shuncheng He, Hongchang Zhang, Xiangyang Ji
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 |
|---|---|---|---|
| Use SPD synergy-pattern pseudo-reward instead of DIAYN/WURL state-diversity rewards | QMIX unsupervised MARL · MPE SimpleTag | Normalized proximal discrepancy of synergy patterns: (4.0 ± 2.7) × 10−2 -> 1.0 ± 0.1 | Improved |
| Use SPD instead of DIAYN/WURL on state identifiability | QMIX unsupervised MARL · MPE SimpleTag | Discriminator success rate (%): 59.9 ± 0.9 -> 58.6 ± 0.7 | No clear effect |
| Initialize downstream QMIX with SPD-pretrained model instead of training QMIX from scratch | QMIX · Google Research Football 3_vs_1 with keeper, four downstream tasks | winning rate | Improved |
| Initialize QMIX with WURL or DIAYN pretrained model | QMIX · GRF downstream tasks | winning rate | No clear effect |
| Scale pseudo-reward factor beta_r over 0.1, 1.0, 10.0 (default 10.0) (ablation) | SPD · MPE SimpleTag | proximal discrepancy of synergy patterns | Mixed |
| Decompose episode return into per-time-step pseudo-rewards (vs sparse final-step return) (ablation) | SPD · MPE SimpleTag | proximal discrepancy of synergy patterns | Improved |
| Add optimal-transport matching rho* instead of rho(t)=t (ablation) | SPD · MPE SimpleTag | proximal discrepancy of synergy patterns | Improved |
| Increase Kuhn-Munkres iterations from 25 to 100 (ablation) | SPD · MPE SimpleTag | proximal discrepancy of synergy patterns: 25 -> 100 | Improved |
Closest papers by meaning
- Towards Principled Unsupervised Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Efficient Heterogeneous Exploration With Mutual Policy Divergence Maximization for Multiagent Reinforcement Learning (TPAMI 2026 · similarity 0.88)
- Wasserstein Unsupervised Reinforcement Learning (AAAI 2022 · similarity 0.88)
- Measuring Mutual Policy Divergence for Multi-Agent Sequential Exploration (NeurIPS 2024 · similarity 0.88)
- Multi-agent cooperation through learning-aware policy gradients (ICLR 2025 · similarity 0.86)
- Maximum Entropy Heterogeneous-Agent Reinforcement Learning (ICLR 2024 · similarity 0.86)
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.
