Changdae Oh
3 indexed papers
Publications per year
Top categories
Frequent co-authors
Research Timeline
The paper proposes Hide-and-Seek, a novel framework that localizes failure signals in VLA model execution by treating failure detection as a coarsely supervised learning problem using contrastive objectives.
The paper proposes Hide-and-Seek, a novel framework that localizes failure signals in VLA model execution by treating failure detection as a coarsely supervised learning problem using contrastive objectives.
The paper shows that reinforcement learning post-training can be used to create effective step-level scoring without dedicated reward model training.
Papers
Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents
Changdae Oh, Wendi Li, Seongheon Park, Samuel Yeh +2 more
The paper shows that reinforcement learning post-training can be used to create effective step-level scoring without dedicated reward model training.