2 citations · 5 across the 10 of their papers we have counts for
8 papers · 1 filter
When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR
Yuchun Miao, Sen Zhang, Yuqi Zhang +4
Reinforcement Learning with Verifiable Rewards (RLVR) has become the dominant paradigm for advanced reasoning in Large Language Models (LLMs), but rollout samples are expensive to…
Information-Theoretic Reward Modeling for Stable RLHF: Detecting and Mitigating Reward Hacking
Yuchun Miao, Liang Ding, Sen Zhang +3
Despite the success of Reinforcement Learning from Human Feedback (RLHF) in aligning language models with human values, reward hacking-or reward over-optimization-remains a major c…
The Energy Loss Phenomenon in RLHF: A New Perspective on Mitigating Reward Hacking
Yuchun Miao, Sen Zhang, Liang Ding +3
This work identifies the Energy Loss Phenomenon in Reinforcement Learning from Human Feedback (RLHF) and its connection to reward hacking. Specifically, energy loss in the final la…
Aligning Few-Step Diffusion Models with Dense Reward Difference Learning
Ziyi Zhang, Li Shen, Sen Zhang +6
Few-step diffusion models enable efficient high-resolution image synthesis but struggle to align with specific downstream objectives due to limitations of existing reinforcement le…
Offline Behavior Distillation
Shiye Lei, Sen Zhang, Dacheng Tao
Massive reinforcement learning (RL) data are typically collected to train policies offline without the need for interactions, but the large data volume can cause training inefficie…
Towards Theoretical Understandings of Self-Consuming Generative Models
Shi Fu, Sen Zhang, Yingjie Wang +2
This paper tackles the emerging challenge of training generative models within a self-consuming loop, wherein successive generations of models are recursively trained on mixtures o…
InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling
Yuchun Miao, Sen Zhang, Liang Ding +3
Despite the success of reinforcement learning from human feedback (RLHF) in aligning language models with human values, reward hacking, also termed reward overoptimization, remains…
Confronting Reward Overoptimization for Diffusion Models: A Perspective of Inductive and Primacy Biases
Ziyi Zhang, Sen Zhang, Yibing Zhan +3
Bridging the gap between diffusion models and human preferences is crucial for their integration into practical generative workflows. While optimizing downstream reward models has…