Paper summary
Paper summary: Recursive Multi-Agent Systems
RecursiveMAS treats a whole multi-agent system as one recursive computation in latent space, linking agents with a light RecursiveLink module and training them jointly; it reports +8.3% accuracy and 1.2–2.4× faster inference.
- Instead of agents exchanging text, they pass latent states around a loop, which is the looped-model idea applied to a team of agents.
- The whole system is co-optimised with gradient-based credit assignment across recursion rounds.