Ting Yao

Learning More with Less: A Dynamic Dual-Level Down-Sampling Framework for Efficient Policy Optimization

From Answer to Think: Multidimensional Supervision of Reasoning Process for LLM Optimization

CAPO: Towards Enhancing LLM Reasoning through Generative Credit Assignment

From Faithfulness to Correctness: Generative Reward Models that Think Critically

Smooth Gradients, Stable Learning: Logits Convexity for Reinforcement Learning

Discriminative Policy Optimization for Token-Level Reward Models