Research map / Multi-agent systems and games / Cooperative multi-agent RL
Learning Agent Communication under Limited Bandwidth by Message Pruning
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 adaptive message gating to ACML | ACML · Traffic control (simple, moderate, and complex) | reward: -37.6 -> -14.6; -103.5 -> -69.4; -2824.9 -> -2101.1 | Helped |
| Add adaptive message gating to ACML | ACML · Simple and moderate routing; simple and complex WAPC | reward: 0.317 -> 0.288; 0.263 -> 0.239; 0.665 -> 0.610; 0.480 -> 0.411 | Hurt |
| Add adaptive message gating to ACML | ACML · Traffic control, routing, and WAPC | message: 100.0% -> 23.9%; 100.0% -> 29.8%; 100.0% -> 25.8%; 100.0% -> 33.6%; 100.0% -> 27.9%; 100.0% -> 41.9%; 100.0% -> 37.7% | Helped |
| Add gating to CommNet | CommNet · Traffic control, routing, and WAPC | message: 100.0% -> 22.8%; 100.0% -> 34.3%; 100.0% -> 29.3%; 100.0% -> 35.2%; 100.0% -> 21.7%; 100.0% -> 19.8%; 100.0% -> 53.1%; 1 | Helped |
| Add gating to AMP | AMP · Traffic control, routing, and WAPC | message: 100.0% -> 18.6%; 100.0% -> 23.7%; 100.0% -> 19.1%; 100.0% -> 46.7%; 100.0% -> 35.0%; 100.0% -> 81.7%; 100.0% -> 57.2%; 1 | Helped |
| Replace the ACML message coordinator with mean aggregation (ablation) | ACML · Traffic control, routing, and WAPC | reward: -37.6 -> -32.9; -103.5 -> -96.1; -2824.9 -> -2661.5; 0.317 -> 0.321; 0.263 -> 0.267; 0.665 -> 0.673; 0.480 -> 0.493 | Helped |
| Replace the ACML message coordinator with attention aggregation (ablation) | ACML · Traffic control, routing, and WAPC | reward: -37.6 -> -24.5; -103.5 -> -91.8; -2824.9 -> -2359.7; 0.317 -> 0.329; 0.263 -> 0.271; 0.665 -> 0.689; 0.480 -> 0.506 | Helped |
| Use a fixed threshold targeting less than 30% message pruning (ablation) | Gated-ACML · Simple and moderate routing | reward: -8.46%; -7.03%; -13.59%; -12.14%; -4.88%; -3.25% | Helped |
| Use a fixed threshold targeting 70–90% message pruning (ablation) | Gated-ACML · Simple and moderate routing | reward: 9.27%; 10.06%; 14.01%; 13.25%; 18.60%; 19.50% | Hurt |
| Prune all messages using a fixed threshold (ablation) | Gated-ACML · Simple and moderate routing | reward: 59.35%; 65.42% | Hurt |
Closest papers by meaning
- Multi-Agent Reinforcement Learning with Communication-Constrained Priors (NeurIPS 2025 · similarity 0.87)
- Deep Reinforcement Learning for Communication Networks (AAAI 2024 · similarity 0.86)
- Cut the Crap: An Economical Communication Pipeline for LLM-based Multi-Agent Systems (ICLR 2025 · similarity 0.86)
- RGMComm: Return Gap Minimization via Discrete Communications in Multi-Agent Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Learning Efficient and Robust Multi-Agent Communication via Graph Information Bottleneck (AAAI 2024 · similarity 0.86)
- Efficient Communication in Multi-Agent Reinforcement Learning with Implicit Consensus Generation (AAAI 2025 · 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.