| name | mute-communication-unlearning-multi-agent |
| description | Return-preserving communication unlearning for efficient multi-agent coordination under bandwidth constraints. Enables MARL agents to selectively forget inter-agent communications while preserving coordination returns, addressing the trade-off between communication sparsity and cooperative performance. Activation: MUTE, communication unlearning, MARL, bandwidth constraints, cooperative games, inter-agent communication, partial observability. |
| metadata | {"arxiv_id":"2607.03473","published":"2026-07-03","authors":"Rui Zuo, Qinwei Huang, Mingyang Li, Zhenhang Zhang, Simon Khan, Qinru Qiu","tags":["mute","communication-unlearning","marl","bandwidth-constraints","cooperative-games","inter-agent-communication","partial-observability"]} |
MUTE: Return-Preserving Communication Unlearning for Efficient Multi-Agent Coordination
Overview
Inter-agent communication is critical for coordinating Multi-Agent Reinforcement Learning (MARL) agents under partial observability to perform effectively in cooperative games; however, real-world bandwidth constraints demand sparse interactions. Prior approaches primarily address this trade-off by designing communication protocols. MUTE introduces a complementary approach: selectively unlearning (forgetting) communications while preserving coordination returns.
Key Problem
Communication Efficiency in MARL
- Agents in cooperative games need to communicate to coordinate under partial observability
- Real-world bandwidth constraints require sparse communication
- Traditional approach: design more efficient communication protocols
- MUTE's approach: selectively forget communications that don't contribute to returns
Key Innovations
Communication Unlearning
- Selectively removes learned communication behaviors that are redundant or low-impact
