(ICLR’2025) RMB: Comprehensively Benchmarking Reward Models in LLM Alignment

Enyu Zhou, Guodong Zheng, Binghai Wang, Zhiheng Xi, Shihan Dou, Rong Bao, Wei Shen, Limao Xiong, Jessica Fan, Yurong Mou, Rui Zheng, Tao Gui, Qi Zhang, Xuanjing Huang

• Proposed a fine-grained benchmarking framework for reward models, covering pairwise and Best-of-N paradigms, and demonstrated the strong correlation between reward model evaluation and downstream alignment tasks.

(AAAI’2025) Alleviating Shifted Distribution in Human Preference Alignment through Meta-Learning

Shihan Dou, Yan Liu, Enyu Zhou, Tianlong Li, Haoxiang Jia, Limao Xiong, Xin Zhao, Junjie Ye, Rui Zheng, Tao Gui, Qi Zhang, Xuanjing Huang

• Proposed a meta-learning method to address distribution shifts in RLHF by alternating optimization of the reward model to adapt to shifted samples and distributions.

(ACL’2024) LoRAMoE: Alleviating world knowledge forgetting in large language models via MoE-style plugin

Shihan Dou*, Enyu Zhou*, Yan Liu, Songyang Gao, Wei Shen, Limao Xiong, Yuhao Zhou, Xiao Wang, Zhiheng Xi, Xiaoran Fan, Shiliang Pu, Jiang Zhu, Rui Zheng, Tao Gui, Qi Zhang, Xuan-Jing Huang

• Explored challenges of world knowledge forgetting during large-scale fine-tuning and proposed LoRAMoE, an architecture combining LoRA and MoE to mitigate knowledge conflicts and enhance multitask capabilities of LLMs.

(EMNLP’2023) RealBehavior: A Framework for Faithfully Characterizing Foundation Models’ Human-like Behavior Mechanisms

Enyu Zhou, Rui Zheng, Zhiheng Xi, Songyang Gao, Xiaoran Fan, Zichu Fei, Jingting Ye, Tao Gui, Qi Zhang, Xuanjing Huang

• Developed a framework for faithfully characterizing LLM human-like behaviors based on psychometric theories, establishing automatic testing processes and investigating the effects of alignment training.