Learning Transferable Topology Priors for Multi-Agent LLM Collaboration Across Domains

In three linesTopoPrior learns reusable collaboration structures for multi-agent LLM systems. Using a conditional variational graph framework, it captures structural regularities across domains offline, reducing online search cost and inference-time token consumption while remaining compatible with existing topology-evolution backbones.

Summary generated by Claude — human-verified