Xiyan Fu

A learned "Token-to-Go" progress meter enables multi-turn RL agents to effectively learn from sparse rewards, boosting performance by up to 7.2 points without increasing training costs.

Machine translation benchmarks have functionally saturated, but pairing human-authored failure cases with deterministic verification rules reveals critical multimodal blind spots that automated metrics consistently miss.

Leveraging hidden states from reward models can boost RLHF performance by over 6% on challenging benchmarks, transforming how we utilize reward signals.

RL can unlock better compositional generalization than supervised fine-tuning by directly optimizing for correct outcomes, especially on complex tasks where supervised models overfit.