OMAC: A Broad Optimization Framework for LLM-Based Multi-Agent Collaboration

2025-05-17

AI Summary

Existing LLM-driven multi-agent systems (MAS) lack systematic optimization strategies, particularly for jointly refining functional capabilities and collaborative structures. Method: This paper introduces the first comprehensive framework featuring five orthogonal optimization dimensions—structure, role, communication, knowledge, and execution—and supports both single-dimension tuning and cross-dimensional joint search. We propose a novel semantic initializer–contrastive discriminator co-design to enable contrastive learning–based collaborative policy optimization, and develop a multi-dimensional joint search algorithm that unifies modeling across all five dimensions. Contribution/Results: Evaluated on code generation, arithmetic reasoning, and general reasoning tasks, our method consistently outperforms state-of-the-art approaches, achieving significant improvements in task completion rate and output quality. Empirical results demonstrate the framework’s strong generalizability, effectiveness, and scalability across diverse MAS configurations and application domains.

Technology Category

• Multiagent Systems: Multiagent Learning
• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Agentic search

Abstract

Agents powered by advanced large language models (LLMs) have demonstrated impressive capabilities across diverse complex applications. Recently, Multi-Agent Systems (MAS), wherein multiple agents collaborate and communicate with each other, have exhibited enhanced capabilities in complex tasks, such as high-quality code generation and arithmetic reasoning. However, the development of such systems often relies on handcrafted methods, and the literature on systematic design and optimization of LLM-based MAS remains limited. In this work, we introduce OMAC, a general framework designed for holistic optimization of LLM-based MAS. Specifically, we identify five key optimization dimensions for MAS, encompassing both agent functionality and collaboration structure. Building upon these dimensions, we first propose a general algorithm, utilizing two actors termed the Semantic Initializer and the Contrastive Comparator, to optimize any single dimension. Then, we present an algorithm for joint optimization across multiple dimensions. Extensive experiments demonstrate the superior performance of OMAC on code generation, arithmetic reasoning, and general reasoning tasks against state-of-the-art approaches.

Problem

• Systematic design and optimization of LLM-based multi-agent systems
• Handcrafted methods limitation in multi-agent collaboration frameworks
• Joint optimization across multiple dimensions for enhanced agent performance

Innovation

• General framework for optimizing LLM-based MAS
• Five key dimensions for agent and structure optimization
• Joint optimization algorithm across multiple dimensions