Yong Lin  Postdoctoral Fellow Princeton Language and Intellegence New Jersey [Google Scholar] [Github] E-mail: yl7690 [at] princeton [dot] edu

Short Bio

I am a postdoctoral fellow at Princeton, jointly hosted by Princeton Language and Intelligence and Chi Jin. I did my PhD study in Tong Zhang's group at the Hong Kong University of Science and Technology (HKUST).

My research focuses on the trustworthiness and applications of machine learning. Key areas of interest include:

• (1) Alignment of AI systems, particularly Large Language Models (LLMs), to prioritize traits like helpfulness, harmlessness, and honesty. This entails ensuring that LLMs not only align with human preferences but also minimize the occurrence of false or misleading information, commonly known as "hallucinations". Recently, I have a strong interest in Reinforcement Learning with Human Preferences (RLHF), which is a promising way to achieve this goal.
• (2) Out-of-Distribution Generalization, such as enabling an autonomous driving system trained on city roads to navigate country roads; ensuring AI diagnostic systems trained on data from one hospital can reliably predict patients from another hospital.

Before starting my PhD studies, I served as a Senior Machine Learning Engineer at Alibaba, a prominent company in China (which open-sourced Qwen series models). I experienced firsthand the impressive capabilities of machine learning and developed industrial-level applications. Concurrently, I gained insights into the inherent challenges and instability of deep models in industrial settings.

I won the Outstanding Paper Award of NAACL (2024). I was also an awardee of Apple AI/ML PhD fellowship (2023) and Hong Kong PhD fellowship (2020).

Selected Papers

(* denotes equal contribution.)

Pre-prints

• Yong Lin*, Chen Liu*, Chenlu Ye*, Qing Lian, Yuan Yao, Tong Zhang.
  Optimal Sample Selection Through Uncertainty Estimation and Its Application in Deep Learning.
  JMLR in submission.
  
• Yifan Hao*, Yong Lin*, Difan Zou, Tong Zhang.
  On the Benefits of Over-parameterization for Out-of-Distribution Generalization.
  Annals of Statistics in submission.
  

Publications

• Qizhou Wang*, Yong Lin*, Yongqiang Chen*, Ludwig Schmidt, Bo Han, Tong Zhang
  Do CLIPs Always Generalize Better than ImageNet Models?
  NeurIPS 2024.
  
• Yong Lin*, Hangyu Lin*, Wei Xiong*, Shizhe Diao*,[+8 authors], Han Zhao , Nan Jiang, Heng Ji, Yuan Yao, and Tong Zhang.
  Mitigating the Alignment Tax of RLHF.
  ENMLP 2024. [ code ]
  
• Yong Lin*, Skyler Seto*, Maartje ter Hoeve, Katherine Metcalf, Barry-John Theobald, Xuan Wang, Yizhe Zhang, Chen Huang, Tong Zhang
  On the Limited Generalization Capability of the Implicit Reward Model Induced by Direct Preference Optimization.
  ENMLP 2024 Findings.
  
• Haoxiang Wang*, Yong Lin*, Wei Xiong*, Rui Yang, Shizhe Diao, Shuang Qiu, Han Zhao, Tong Zhang
  Arithmetic Control of LLMs for Diverse User Preferences: Directional Preference Alignment with Multi-Objective Rewards.
  ACL 2024.
  
• Hanning Zhang*, Shizhe Diao*, Yong Lin*, Yi R. Fung, Qing Lian, Xingyao Wang, Yangyi Chen, Heng Ji, Tong Zhang.
  R-tuning: Teaching large language models to refuse unknown questions.
  NAACL 2024 [Outstanding Paper Award, 6/2434 = 0.25%] .
  
• Yong Lin*, Lu Tan*, Yifan Hao*, Honam Wong, Hanze Dong, Weizhong Zhang, Yujiu Yang, Tong Zhang.
  Spurious Feature Diversification Improves Out-of-distribution Generalization.
  ICLR 2024.
  
• Damien Teney, Yong Lin, Seong Joon Oh, Ehsan Abbasnejad.
  Id and ood performance are sometimes inversely correlated on real-world datasets.
  NeurIPS 2023 [Spotlight].
  
• Rui Yang, Yong Lin, Xiaoteng Ma, Hao Hu, Chongjie Zhang, Tong Zhang.
  What Is Essential for Unseen Goal Generalization of Offline Goal-conditioned RL?
  ICML 2023
• Yong Lin*, Renjie Pi*, Weizhong Zhang, Xiaobo Xia, Jiahui Gao, Xiao Zhou, Tongliang Liu, Bo Han.
  A Holistic View of Noise Transition Matrix in Deep Learning and Beyond?
  ICLR 2023 [Spotlight].
• Yong Lin, Shengyu Zhu, Lu Tan, Peng Cui.
  ZIN: When and How to Learn Invariance by Environment Inference?
  NeurIPS 2022 [Spotlight].
• Yong Lin*, Hanze Dong*, Hao Wang, Tong Zhang.
  Bayesian Invariant Risk Minimization
  CVPR 2022 [Oral].
• Xiao Zhou*, Yong Lin*, Weizhong Zhang*, Tong Zhang.
  Sparse Invariant Risk Minimization.
  ICML 2022.
• Xiao Zhou*, Yong Lin*, Renjie Pi*, Weizhong Zhang, Renzhe Xu, Peng Cui, Tong Zhang.
  Model Agnostic Sample Reweighting for Out-of-Distribution Learning.
  ICML 2022.
• Yong Lin*, Qing Lian* and Tong Zhang.
  An Empirical Study of Invariant Risk Minimization on Deep Models.
  ICML2021 workshop on UDL.
• Yong Lin, Zheng Xu.
  Cable sheath loss reduction strategy research based on the coupled linemodel.
  IEEE Transactions On Power Delivery.

Selected Awards

• Outstanding Paper Award of NAACL 2024
• 2023 Apple Scholars in AI/ML PhD fellowship (22 awardees all over the world).
• Hong Kong PhD Fellowship.
• National Scholarship * 3 (1.8%, by China's Ministry of Education), 2010, 2011 and 2015.
• Outstanding Graduate of Zhejiang Province, 2013.

Experiences

• Princeton University , Postdoc Fellow, 2024.9-now.
• The Hong Kong University of Science and Technology , PhD Student, 2020 - 2024.
• Alibaba , Senior Machine Learning Engineer, 2016 - 2020.
• Zhejiang University , Bachelor and Master Student (Ranking 1/207), 2009 - 2016.