Announcement
RecursiveMAS accepted at NeurIPS 2026
A post announcing the acceptance describes RecursiveMAS as a team of agents that pass and refine latent states across rounds of collaboration, with each agent acting as a layer.
TLDR
The post says RecursiveMAS pairs that agent-team design with inner–outer loop training so the agents learn to collaborate through recursion. Its stated aim is stronger reasoning and faster inference.
Combined views
2.1K
4 Sources, first seen ago
