Research map / Multi-agent systems and games / Cooperative multi-agent RL
Evolutionary Population Curriculum for Scaling 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 |
|---|---|---|---|
| add population curriculum plus evolutionary selection (EPC) over vanilla population curriculum (ablation) | MADDPG · Grassland scale 24-16 and Food Collection scale 24 | team/wolf/sheep reward vs baselines: Food Collection 59.47 vs 49.998 (vanilla-PC); Grassland wolf 61.42 vs 47.64 | Helped |
| add evolutionary selection to the population curriculum (EPC vs vanilla-PC) (ablation) | MADDPG · Adversarial Battle scale 16-16 | team 1 reward: 189.70 vs 109.44 | Helped |
| population curriculum with progressive doubling of agents (vanilla-PC) vs training from scratch with population-invariant architecture (ablation) | MADDPG (Att-MADDPG architecture) · Grassland scale 24-16 | wolf reward vs EPC sheep: 47.64 vs 32.89 | Helped |
| population curriculum (vanilla-PC) vs training from scratch (ablation) | MADDPG (Att-MADDPG architecture) · Grassland scale 6-4 | wolf reward vs EPC sheep: 2.001 vs 2.5905 (Att-MADDPG) | No clear effect |
| replace MADDPG critic/policy with population-invariant attention architecture (ablation) | MADDPG · Food Collection scale 24 | team reward: 33.435 vs 10.346 | Helped |
| use mean-field approximation (Yang et al. 2018) instead of MADDPG | MADDPG · Food Collection scale 24 | team reward: 7.83 vs 10.35 (MADDPG) | Hurt |
| use mean-field approximation instead of MADDPG | MADDPG · Grassland and Adversarial Battle | reward: little differences between mean-field and MADDPG | No clear effect |
| add evolutionary selection (EPC vs vanilla-PC) to reduce variance across seeds (ablation) | MADDPG · Grassland, Adversarial Battle, Food Collection (second stage) | variance of agent scores over 3 seeds: qualitative: variance much smaller than vanilla-PC | Helped |
Closest papers by meaning
- Learning Progress Driven Multi-Agent Curriculum (ICML 2025 · similarity 0.88)
- Controlling Behavioral Diversity in Multi-Agent Reinforcement Learning (ICML 2024 · similarity 0.87)
- HyperMARL: Adaptive Hypernetworks for Multi-Agent RL (NeurIPS 2025 · similarity 0.86)
- Accelerate Multi-Agent Reinforcement Learning in Zero-Sum Games with Subgame Curriculum Learning (AAAI 2024 · similarity 0.86)
- PORTAL: Automatic Curricula Generation for Multiagent Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Two-Stage Evolutionary Reinforcement Learning for Enhancing Exploration and Exploitation (AAAI 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.