Research map / Multi-agent systems and games / Cooperative multi-agent RL
Value-Decomposition Multi-Agent Actor-Critics
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 VDAC-mix instead of other actor-critic methods | Actor-critic methods · SMAC maps | median win rate | Helped |
| Use VDAC-mix instead of QMIX under A2C training (ablation) | VDAC-mix and QMIX · 3s5z | median test win rate: 71% | Helped |
| Use VDAC-mix instead of QMIX under A2C training (ablation) | VDAC-mix and QMIX · 2s_vs_1sc | win rate | Helped |
| Use TD-advantage policy gradients instead of COMA gradients (ablation) | Naive critic versus COMA · SMAC maps | win rate | Helped |
| Add a central critic with access to extra state information (ablation) | IAC · SMAC maps | win rate | Helped |
| Add monotonic state-value factorization by summing local state values (ablation) | IAC · SMAC maps | median win rate | Helped |
| Use summed local state values instead of a central critic (ablation) | Naive critic · SMAC maps | win rate | No clear effect |
| Use nonlinear value decomposition instead of linear value decomposition (ablation) | VDAC-mix · SMAC maps | median win rate | Helped |
| Add access to extra state information (ablation) | VDAC-sum, yielding VDAC-mix (linear) · SMAC maps | win rate | Helped |
| Use COMA gradients instead of TD-advantage gradients (ablation) | Multi-agent actor-critic · SMAC maps | training stability | Hurt |
| Use QMIX under A2C training instead of VDAC-mix (ablation) | QMIX · 2s_vs_1sc | training stability | Hurt |
| Use VDAC-mix instead of QMIX under A2C training (ablation) | VDAC-mix and QMIX · Easier SMAC maps | win rate | No clear effect |
Closest papers by meaning
- QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (ICML 2018 · similarity 0.90)
- Value Function Decomposition for Iterative Design of Reinforcement Learning Agents (NeurIPS 2022 · similarity 0.89)
- FOP: Factorizing Optimal Joint Policy of Maximum-Entropy Multi-Agent Reinforcement Learning (ICML 2021 · similarity 0.89)
- Weighted QMIX: Expanding Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (NeurIPS 2020 · similarity 0.88)
- Q-value Path Decomposition for Deep Multiagent Reinforcement Learning (ICML 2020 · similarity 0.88)
- Actor-Free Continuous Control via Structurally Maximizable Q-Functions (NeurIPS 2025 · 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.