AI summaryⓘ
The authors focus on improving communication between multiple learning agents working together when they can’t see the whole environment. They point out that previous methods tried to reduce communication by using general statistical measures, which might keep messages that don’t actually help the team’s success. To fix this, they introduce MUTE, a way to identify and stop sending messages that don’t contribute much to the group’s performance. Their method carefully balances sending fewer messages while keeping the team’s overall effectiveness high, and they prove this approach won’t greatly hurt performance. They tested MUTE in complex tasks and showed it can cut communication by 80-90% without losing skill compared to other methods.
Multi-Agent Reinforcement LearningPartial ObservabilityInter-agent CommunicationBandwidth ConstraintsMessage UnlearningCounterfactual Message ValueAttention-based EstimatorPolicy SparsificationPerformance BoundCooperative Games
Authors
Rui Zuo, Qinwei Huang, Mingyang Li, Zhenhang Zhang, Simon Khan, Qinru Qiu
Abstract
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 optimizing information-theoretic surrogates. We argue that these statistical proxies are fundamentally misaligned with the true objective: a message can be highly informative yet irrelevant to the joint return of the task. In this work, we propose Message Unlearning for Targeted Efficiency (MUTE), a framework that views communication reduction as a value-guided machine unlearning problem. MUTE rigorously quantifies the Counterfactual Message Value using an attention-based estimator, and systematically unlearns the transmission of low-value messages from a policy trained without any communication constraints. This is achieved through a dual-objective mechanism that enforces communication sparsity while preserving the return of the original joint policy. We derive a theoretical upper bound on the performance gap induced by this sparsification, guaranteeing controlled return degradation. We also empirically evaluate MUTE on various complex multi-agent environments, achieving 80% to 90% bandwidth reduction while maintaining performance comparable to state-of-the-art baselines.