Research map / Multi-agent systems and games / Cooperative multi-agent RL
More Centralized Training, Still Decentralized Execution: Multi-Agent Conditional Policy Factorization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| introduce dependent local policies during centralized training | MACPF · matrix game | optimal policy discovery: 60% for QPLEX, 20% for QTRAN, and 40% for FOP | Helped |
| introduce dependency among agents during training (ablation) | MACPF · SMAC, four ablation maps | win rate | Helped |
| use the learned independent policy instead of the dependent policy (ablation) | MACPF · SMAC | win rate | No clear effect |
| use a hypernetwork mixer with non-linearity (ablation) | MACPF · SMAC easy and hard maps | win rate | No clear effect |
| use a hypernetwork mixer with non-linearity (ablation) | MACPF · SMAC super-hard maps | win rate | Hurt |
| use weighted summation with actions as mixer inputs (ablation) | MACPF · SMAC maps | win rate | Hurt |
| use dependent policies and explicit agent dependency | MACPF · SMAC | win rate | Helped |
| use MACPF rather than the strongest listed baseline | MACPF · MPE Simple Spread | average reward per episode: -122.50 -> -118.24 | Helped |
| use MACPF rather than the strongest listed baseline | MACPF · MPE Formation Control | average reward per episode: -15.84 -> -15.79 | No clear effect |
| use MACPF rather than the strongest listed baseline | MACPF · MPE Line Control | average reward per episode: -19.78 -> -19.60 | No clear effect |
| use MACPF rather than FOP | MACPF · SMAC final performance | win rate: The win rate of MACFP is higher than or equivalent to the best baseline in 16 out of 18 maps; FOP, 5/18. | Mixed |
Closest papers by meaning
- Revisiting Some Common Practices in Cooperative Multi-Agent Reinforcement Learning (ICML 2022 · similarity 0.91)
- Communication-Efficient Actor-Critic Methods for Homogeneous Markov Games (ICLR 2022 · similarity 0.91)
- Learning to Act in Decentralized Partially Observable MDPs (ICML 2018 · similarity 0.90)
- AgentMixer: Multi-Agent Correlated Policy Factorization (AAAI 2025 · similarity 0.89)
- Cooperative Policy Agreement: Learning Diverse Policy for Offline MARL (AAAI 2025 · similarity 0.89)
- Backpropagation Through Agents (AAAI 2024 · 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.