Yongdong Zhang

Unsupervised Reinforcement Learning with Verifiable Rewards via First Repeat Criterion

UpSafe℃: Upcycling for Controllable Safety in Large Language Models

Video-LevelGauge: Investigating Contextual Positional Bias in Video Language Models.

INPO: Image-based Negative Preference Optimization for Concept Erasure in Text-to-Image Diffusion Models

Beyond Unified Directions: Context-adaptive Representation Steering for LLM Safety Alignment

Leveraging Importance Sampling to Detach Alignment Modules from Large Language Models

Leveraging robust optimization for llm alignment under distribution shifts