Research map / Multi-agent systems and games / Cooperative multi-agent RL
Scaling Multi-Agent Reinforcement Learning with Selective Parameter Sharing
Filippos Christianos, Georgios Papoudakis, Muhammad A Rahman, Stefano V Albrecht
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 |
|---|---|---|---|
| Selectively share parameters among clustered agents instead of giving each agent an independent policy | A2C MARL · BPS | sum of returns: -189.99 -> -179.69; -385.18 -> -369.69; -401.09 -> -371.99; -388.21 -> -378.50 | Improved |
| Selectively share parameters among clustered agents instead of giving each agent an independent policy | A2C MARL · BPS-h | sum of returns: -187.92 -> -189.93; -398.11 -> -373.92 | Mixed |
| Selectively share parameters among clustered agents instead of using SAPI | A2C MARL · BPS-h (3) | sum of returns: -4085.51 -> -2522.61 | Improved |
| Selectively share parameters among clustered agents instead of using independent policies | A2C MARL · C-RWARE | sum of returns: 2.93 -> 7.03; 0.28 -> 20.88 | Improved |
| Selectively share parameters among clustered agents instead of using SAPI | A2C MARL · C-RWARE (3) | sum of returns: 21.30 -> 32.27 | Improved |
| Selectively share parameters among clustered agents instead of independent policies | A2C MARL · LBF | sum of returns: 0.91 -> 0.97 | Improved |
| Share parameters selectively rather than using SAPI | A2C MARL · MMM2 | sum of returns: 168.39 -> 179.45 | Improved |
| Use fully shared parameters instead of independent policies | A2C MARL · MMM2 | sum of returns: 72.32 -> 128.75 | Improved |
| Use fully shared parameters instead of independent policies | A2C MARL · BPS and C-RWARE | sum of returns: -189.99 -> -612.18; 2.93 -> 0.57 | Hurt |
| Use independent policies instead of parameter sharing | A2C MARL · C-RWARE (2) and (3) | sum of returns: 20.88 -> 0.28; 32.27 -> 0.33 | Hurt |
| Scale up SAPI network width with the number of BPS-h colours (ablation) | SAPI · BPS-h | learning success | No clear effect |
| Form fewer clusters than needed (ablation) | OPS · C-RWARE | sum of returns | Hurt |
Closest papers by meaning
- GradPS: Resolving Futile Neurons in Parameter Sharing Network for Multi-Agent Reinforcement Learning (ICML 2025 · similarity 0.90)
- HyperMARL: Adaptive Hypernetworks for Multi-Agent RL (NeurIPS 2025 · similarity 0.88)
- Kaleidoscope: Learnable Masks for Heterogeneous Multi-agent Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Mixtures of Experts Unlock Parameter Scaling for Deep RL (ICML 2024 · similarity 0.87)
- Learning to Share: Selective Memory for Efficient Parallel Agentic Systems (ICML 2026 · similarity 0.86)
- Cautiously-Optimistic Knowledge Sharing for Cooperative Multi-Agent Reinforcement Learning (AAAI 2024 · similarity 0.85)
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.
