Chinese Academic of Science Institute of Automation

ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning

From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Space