Research map / Multi-agent systems and games / Cooperative multi-agent RL
ADAPT: Adaptive Decentralized Architecture with Perception-Aligned Training for Structural Generalization in Multi-Agent RL
Zhixiang Zhang, Shuo Chen, Yexin Li, Feng 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 |
|---|---|---|---|
| select actions with FINE instead of majority voting among source policies (ablation) | ADAPT · SMAC zero-shot 5m to 7m | test win rate (%): 73.44 -> 85.76 | Improved |
| select actions with FINE instead of majority voting among source policies (ablation) | ADAPT · SMAC zero-shot 3m to 5m | test win rate (%): 43.13 -> 47.16 | Improved |
| select actions with FINE instead of voting (ablation) | ADAPT · MPE zero-shot (12 target maps) | test win rate (%): better on 9 of 12 maps | Mixed |
| select actions with FINE instead of voting (ablation) | ADAPT · MPE 3prey-3predator to 12prey-10predator | test win rate (%): 53.17 -> 73.50 | Improved |
| select actions with FINE instead of voting (ablation) | ADAPT · MPE 2prey-4predator to 6prey-4predator | test win rate (%): 84.90 -> 75.60 | Hurt |
| use ADAPT instead of I2Q, trained from scratch | I2Q · SMAC 5m_vs_6m | test win rate (%): 40.00 -> 86.25 | Improved |
| use ADAPT instead of I2Q, trained from scratch | I2Q · SMAC 2s3z | dead allies: 2.89 -> 1.33 | Improved |
| use ADAPT instead of I2Q, trained from scratch | I2Q · SMAC 5m | test win rate (%): 99.38 -> 98.13 | No clear effect |
| use ADAPT+FINE zero-shot instead of OPT-DTDE | OPT-DTDE · SMAC zero-shot 5m to 7m | test win rate (%): 20.20 -> 85.76 | Improved |
| use ADAPT-share (shared gradients, identical networks) instead of independent networks with FINE | ADAPT · MPE zero-shot 3prey-3predator to 20prey-20predator | test win rate (%): 79.17 -> 63.50 | Hurt |
| ADAPT+FINE vs OPT-CTDE zero-shot after training on 2prey-4predator | ADAPT+FINE · MPE 2prey-4predator source | test win rate (%): exceeds OPT-CTDE on 4 of 6 zero-shot maps | Mixed |
Closest papers by meaning
- Multi-agent In-context Coordination via Decentralized Memory Retrieval (AAAI 2026 · similarity 0.89)
- Towards Generalizable Reinforcement Learning via Causality-Guided Self-Adaptive Representations (ICLR 2025 · similarity 0.88)
- Dr. MAS: Stable Reinforcement Learning for Multi-Agent LLM Systems (NeurIPS 2026 · similarity 0.87)
- Adaptive Collaboration with Humans: Metacognitive Policy Optimization for Multi-Agent LLMs with Continual Learning (ICLR 2026 · similarity 0.87)
- HyperMARL: Adaptive Hypernetworks for Multi-Agent RL (NeurIPS 2025 · similarity 0.86)
- Enabling Adaptive Agent Training in Open-Ended Simulators by Targeting Diversity (NeurIPS 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.
