TopoDIM: One-shot Topology Generation of Diverse Interaction Modes for Multi-Agent Systems

A research team led by Rui Sun has introduced TopoDIM, a novel framework that revolutionizes how LLM-based multi-agent systems communicate. Traditional methods rely on spatio-temporal interaction paradigms requiring sequential, multi-round dialogues that incur high latency and computational costs. TopoDIM addresses this by enabling agents to autonomously construct heterogeneous communication topologies in a single step, drawing inspiration from evaluation and debate mechanisms that enhance collective problem-solving.

Designed for decentralized execution, TopoDIM enhances both adaptability and privacy by eliminating the need for iterative coordination between agents. Experiments demonstrate significant efficiency gains: the framework reduces total token consumption by 46.41% while simultaneously improving average task performance by 1.50% compared to current state-of-the-art methods. This one-shot topology generation approach also shows strong adaptability in organizing communication among heterogeneous agents with different capabilities.

The technical breakthrough lies in moving beyond sequential dialogue patterns to optimized, pre-planned interaction structures. By generating diverse interaction modes upfront—including debate, evaluation, and collaborative patterns—agents can execute tasks more efficiently without the back-and-forth overhead. The framework's code is publicly available, potentially accelerating development of more cost-effective multi-agent applications across various domains.