| 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
- Preserves the cooperative return (team reward) despite reduced communication
- Complementary to existing communication-efficient MARL methods
Return-Preserving Guarantee
- Unlearning process explicitly targets communications that don't affect coordination returns
- Ensures that removing communication doesn't degrade team performance
- Distinguishes between communications that are critical vs. those that are noise
Methodology
- Communication Analysis: Identify which inter-agent messages contribute to returns
- Unlearning Process: Selectively remove non-critical communication behaviors from trained policies
- Return Preservation: Validate that team performance is maintained after unlearning
- Bandwidth Reduction: Measure communication sparsity achieved while preserving returns
Implications
- Communication unlearning as a new paradigm for MARL efficiency
- Complements existing bandwidth-efficient communication protocols
- Demonstrates that trained MARL policies often have redundant communication
- Practical for deploying MARL in bandwidth-constrained real-world environments
Pitfalls
- Unlearning may remove communications that are marginally useful in edge cases
- Return preservation is validated empirically, not with formal guarantees
- Unlearning process itself has computational cost
- May not generalize across different cooperative game structures
- Interaction with existing communication-efficient methods needs further study
Activation Keywords
MUTE, communication unlearning, MARL, bandwidth constraints, cooperative games, inter-agent communication, partial observability, communication sparsity, return preservation, multi-agent coordination
Paper Reference
arXiv:2607.03473 - "MUTE: Return-Preserving Communication Unlearning for Efficient Multi-Agent Coordination" (Jul 2026)