\textsc{MasFACT}: Continual Multi-Agent Topology Learning via Geometry-Aware Posterior Transfer

In three linesMasFACT introduces a geometry-aware posterior transfer framework for LLM-powered multi-agent systems. It addresses topology forgetting by preserving historical collaboration structures when adapting to new tasks, using Fused Gromov-Wasserstein optimal transport and PAC-Bayes-guided conservative posterior adaptation.

Summary generated by Claude — human-verified