MasFACT Framework Mitigates Topology Forgetting in Multi-Agent Systems

MasFACT addresses topology forgetting in LLM-powered multi-agent systems, where adapting to new tasks destroys effective communication structures from previous tasks. The framework uses geometry-aware posterior transfer to enable continual learning of agent topologies across evolving task streams.