Llm agents learn who to talk to for better teamwork and less waste
Learning Who Talks to Whom: Role-Aware Communication Topology Scheduling for Efficient LLM Multi-Agent Collaboration
Multiagent Systems
Summary
Large language models can work together as teams to solve complex problems, but deciding which agents should communicate and when is hard. The authors propose a new method called ACTS that helps these agents dynamically choose who to talk with based on their roles and current tasks. This approach improves team accuracy while reducing unnecessary communication, saving computing resources. Tests on different reasoning and coding tasks show ACTS works better than previous methods.
What this means in practice
- •For software development teams: Enable teams of AI coding assistants to collaborate more accurately while reducing communication overhead during code generation.
- •For ai system integrators: Build multi-agent AI systems with efficient dynamic communication for complex problem solving in domains like mathematics and knowledge reasoning.
Authors
Haibin Li, Wenya Yu, Xuefeng Jiang, Jia Li, Xiao Han, Shuai Su, Wei Wei
Abstract
Multi-agent systems driven by large language models (LLMs) rely on structured communication topologies to facilitate information exchange and collaborative reasoning. However, effectively scheduling these topologies to dynamically determine who talks to whom and when to communicate remains a fundamental challenge in balancing collective reasoning performance against exponential token overhead. Existing topology scheduling approaches either enforce static communication graphs that cannot adapt to dynamic problem-solving states, or overlook agent role complementarity while failing to explicitly incorporate the synthesized graph structure into team-level decision evaluation. To tackle these limitations, we propose ACTS, a role-aware adaptive communication topology scheduling framework for efficient LLM multi-agent collaboration. Orchestrated within a Centralized Training with Decentralized Execution (CTDE) paradigm, ACTS introduces a novel dual-attention mechanism. Specifically, a Role-conditioned Relational Attention module dynamically selects variable-cardinality interaction partners based on agent roles and evolving execution states, while a Topological Attention Mixer explicitly embeds the directed communication topology into team value decomposition for precise credit assignment. Comprehensive evaluations across 9 challenging benchmarks covering code generation, multi-step reasoning, general knowledge, and competitive mathematics demonstrate that ACTS consistently outperforms state-of-the-art baselines, boosting average accuracy by up to 6.19 percentage points while reducing token consumption by over 15%. The project website is available at https://acts-mas.github.io/ACTS/.