Research map / Multi-agent systems and games / Cooperative multi-agent RL
QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning
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 |
|---|---|---|---|
| use monotonic nonlinear mixing of agent values with state-conditioned weights | QMIX · StarCraft II micromanagement maps | test win rate | Helped |
| use joint action-value factorisation instead of independent value learning | VDN versus IQL · StarCraft II micromanagement maps | test win rate | Helped |
| add nonlinear mixing to agent values | QMIX versus VDN · StarCraft II 8m | test win rate | No clear effect |
| remove state-conditioned hypernetworks (ablation) | QMIX · StarCraft II heterogeneous-agent maps | test win rate | Hurt |
| remove the nonlinear mixing hidden layer (ablation) | QMIX · StarCraft II heterogeneous-agent maps | test win rate | Hurt |
| remove state conditioning from nonlinear mixing (ablation) | QMIX · StarCraft II complex scenarios | test win rate | No clear effect |
| add a state-dependent term to VDN's additive value sum (ablation) | VDN · StarCraft II heterogeneous-agent maps | test win rate | Hurt |
| remove nonlinear mixing while retaining state-conditioned weighted summation (ablation) | QMIX · Two-Step Game | final test reward: 8 -> 7 | Hurt |
| remove state-conditioned hypernetworks (ablation) | QMIX · Two-Step Game | final test reward: 8 -> 8 | No clear effect |
| replace nonlinear mixing with VDN's additive decomposition (ablation) | QMIX · Two-Step Game | final test reward: 8 -> 7 | Hurt |
| add a state-dependent term to VDN's additive decomposition (ablation) | VDN · Two-Step Game | final test reward: 7 -> 7 | No clear effect |
| use centralised monotonic factorisation instead of independent Q-learning | QMIX versus IQL · Two-Step Game | final test reward: 7 -> 8 | Helped |
Closest papers by meaning
- Weighted QMIX: Expanding Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (NeurIPS 2020 · similarity 0.95)
- QTRAN: Learning to Factorize with Transformation for Cooperative Multi-Agent Reinforcement Learning (ICML 2019 · similarity 0.91)
- Towards Understanding Cooperative Multi-Agent Q-Learning with Value Factorization (NeurIPS 2021 · similarity 0.91)
- MAVEN: Multi-Agent Variational Exploration (NeurIPS 2019 · similarity 0.91)
- ConcaveQ: Non-monotonic Value Function Factorization via Concave Representations in Deep Multi-Agent Reinforcement Learning (AAAI 2024 · similarity 0.90)
- AgentMixer: Multi-Agent Correlated Policy Factorization (AAAI 2025 · 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.