Research map / Multi-agent systems and games / Cooperative multi-agent RL
Surprise Minimizing Multi-Agent Learning with Energy-based Models
Karush Suri, Xiao Qi Shi, Konstantinos N Plataniotis, Yuri Andrew Lawryshyn
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 |
|---|---|---|---|
| add EMIX energy-based surprise minimization on top of QMIX | QMIX · StarCraft II micromanagement (6 scenarios) | success rate: 93.43 -> 94.90 (3m); 34.42 -> 37.07 (5m_vs_6m) | Improved |
| add EMIX to QMIX on large-opponent scenario | QMIX · so_many_baneling | success rate: 93.35 -> 95.87 | Improved |
| add EMIX to QMIX on 10m_vs_11m | QMIX · 10m_vs_11m | success rate: 70.36 -> 75.35 | Improved |
| use EMIX energy-based surprise instead of SMiRL surprise in QMIX | SMiRL-QMIX · StarCraft II micromanagement | success rate: 0.24 -> 97.22 (3s_vs_4z); 77.85 -> 75.35 (10m_vs_11m) | Mixed |
| add EMIX to QMIX on Predator-Prey punish | QMIX · predator_prey_punish | average return: 14.33 -> 24.17 | Improved |
| add EMIX to QMIX on Predator-Prey hard | QMIX · predator_prey_hard | average return: 8.76 -> 12.34 | Improved |
| add EMIX to QMIX on easy Predator-Prey tasks | QMIX · predator_prey_easy and predator_prey | average return: 40.00 -> 40.00 | No clear effect |
| add energy operator on top of twin surprise approximators (TwinQMIX -> EMIX) (ablation) | TwinQMIX · predator_prey_punish | average return: 20.32 -> 24.17 | Improved |
| add energy operator on top of twin surprise approximators (TwinQMIX -> EMIX) (ablation) | TwinQMIX · predator_prey_hard | average return: 10.19 -> 12.34 | Improved |
| add twin surprise approximators to QMIX (TwinQMIX) (ablation) | QMIX · 3 StarCraft II scenarios | average success rate | Improved |
| use dual (twin) approximators for surprise in EMIX (ablation) | EMIX · 3 StarCraft II scenarios | average success rate | No clear effect |
| add energy operator T to surprise value function (ablation) | TwinQMIX · 5m_vs_6m and so_many_baneling | average success rate | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- QMIX — QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning (ICML 2018) · better on 8 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Principled Unsupervised Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- On Feasible Rewards in Multi-Agent Inverse Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
- AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement Learning (NeurIPS 2026 · similarity 0.84)
- Multi-Agent Learning under Uncertainty: Recurrence vs. Concentration (NeurIPS 2025 · similarity 0.84)
- Foundations of Multi-Agent Learning in Dynamic Environments: Where Reinforcement Learning Meets Strategic Decision-Making (AAAI 2025 · similarity 0.84)
- Decomposing Temporal Equilibrium Strategy for Coordinated Distributed Multi-Agent Reinforcement Learning (AAAI 2024 · similarity 0.84)
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.
