Multi-agent systems improve workflow designs with guided learning
FlowMAS: Learning Multi-Agent Workflow Topology via Information-guided Generative Flow Network
Multiagent Systems
Summary
Designing workflows for multiple interacting agents is hard because existing methods are too slow, too rough, or not good at handling complex connections. The authors propose FlowMAS, which uses a special learning system called Generative Flow Networks to create better workflows step by step, guided by rewards. They add curious exploration to try new designs and smart measures to judge how well the workflows share information. Tests show FlowMAS works better than older methods on several tasks.
What this means in practice
- •For ai system engineers: Design automated multi-agent workflows that learn effective collaboration patterns for complex tasks using reward signals and information efficiency measures.
- •For software architects: Generate scalable and adaptable workflow topologies for distributed systems by leveraging curiosity-driven exploration of structural designs.
Authors
Haitao Wang, Chenjing Liang, Haipeng Zhang, Jiawei Hu, Sicheng Wang, Songzhu Mei, Chenglu Wen, Siqi Shen, Cheng Wang
Abstract
Automated multi-agent systems offer clear advantages over manually designed ones in scalability and adaptability, but existing workflow topology methods still face important limitations. Search-based methods are often computationally expensive, textual-gradient-based methods rely on coarse-grained feedback, and existing generation-based methods are not well suited to discrete workflow topologies with complex dependencies. To address these limitations, we propose FlowMAS, a multi-agent workflow topology method based on Generative Flow Networks (GFlowNets). FlowMAS models workflow generation as reward-guided flow over the topology space and introduces three components: a GFlowNet-based topology generation backbone, a curiosity-driven module for structure-aware exploration, and an information-guided optimization module for evaluating intermediate topologies. Concretely, the curiosity-driven module encourages exploration of structurally novel workflows, while the information-guided module measures both the information contribution and the communication efficiency of different operators to favor more informative and effective collaboration patterns. Experiments on six benchmark datasets with three LLM backbones show that FlowMAS consistently outperforms multiple baselines.