Reproduce “FlowMAS: Learning Multi-Agent Workflow Topology via Information-guided Generative Flow Network” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haitao Wang, Chenjing Liang, Haipeng Zhang, Jiawei Hu, Sicheng Wang, Songzhu Mei, Chenglu Wen, Siqi Shen and 1 more
FlowMAS learns to build directed acyclic workflow graphs for multi-agent systems, adding operators and connections step by step with a GFlowNet. Intermediate rewards guide construction using structural novelty and embedding-based information and channel heuristics. Across six benchmarks and three backbones, it reports improved performance; on HumanEval with Llama-3.1-8B-Instruct, it scores 69.6±1.5 pass@1 versus AFlow’s 65.1±1.4, within the tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.