method · first seen 12 Oct 2026
reinforcement learning from system outcomes
Also: compositional reinforcement learning
Training a policy by rewarding the outcome of a whole multi-model system rather than the policy alone.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Compositional RL: Evaluating and Training LLMs for Collaborative ReasoningPapers · 12 Oct 2026 · Shangbin Feng, Yulia Tsvetkov, Jan Kautz, David Acuna
Latest sources
- Compositional RL: Evaluating and Training LLMs for Collaborative ReasoningPapers · 12 Oct 2026 · Shangbin Feng, Yulia Tsvetkov, Jan Kautz, David Acuna