Wei Hu (胡威) Email: vvh [at] umich [dot] edu [Google Scholar] [Twitter] [LinkedIn]

About

I am a researcher at OpenAI and an assistant professor (on leave) in Computer Science and Engineering at the University of Michigan.

Previously, I was a FODSI postdoc at UC Berkeley, did my PhD in Computer Science at Princeton University advised by Sanjeev Arora, and did my undergrad at Tsinghua University where I was a member of Yao Class.

Group at UMich

Current members

• Pulkit Gopalani (PhD student, co-advised with Gabriel Poesia)

• Zhiwei Xu (PhD student, co-advised with Yixin Wang)

Past members

• Yongyi Yang (PhD 2026. Next: Postdoc at Harvard)

• Yutong Wang (Postdoc co-hosted with Qing Qu. Next: Assistant professor at Illinois Tech)

Papers

α-β indicates alphabetical author order.
* indicates equal contribution.
† indicates equal advising.

• Data-Constrained Language Model Pretraining: Improved Regularization and Scaling Laws
  Zhiwei Xu, Shihao Wu, Hanseul Cho, Wei Hu†, Yixin Wang†
  Manuscript, 2026

• Skill-Aware Data Selection and Fine-Tuning for Data-Efficient Reasoning Distillation
  Lechen Zhang, Yunxiang Zhang, Wei Hu, Lu Wang
  ACL 2026

• Robust Sparse Mean Estimation via Incremental Learning
  Jianhao Ma, Rui Ray Chen, Yinghui He, Salar Fattahi, Wei Hu
  TMLR 2026

• What Happens During the Loss Plateau? Understanding Abrupt Learning in Transformers
  Pulkit Gopalani, Wei Hu
  NeurIPS 2025

• Benign Overfitting in Single-Head Attention
  Roey Magen*, Shuning Shang*, Zhiwei Xu, Spencer Frei, Wei Hu†, Gal Vardi†
  NeurIPS 2025

• Provable Low-Frequency Bias of In-Context Learning of Representations
  Yongyi Yang, Hidenori Tanaka, Wei Hu
  Manuscript, 2025

• RaanA: A Fast, Flexible, and Data-Efficient Post-Training Quantization Algorithm
  Yongyi Yang, Jianyang Gao, Wei Hu
  Manuscript, 2025

• Understanding Deep Representation Learning via Layerwise Feature Compression and Discrimination
  Peng Wang*, Xiao Li*, Can Yaras, Zhihui Zhu, Laura Balzano, Wei Hu, and Qing Qu
  JMLR 2025

• Let Me Grok for You: Accelerating Grokking via Embedding Transfer from a Weaker Model
  Zhiwei Xu*, Zhiyu Ni*, Yixin Wang†, Wei Hu†
  ICLR 2025

• Swing-by Dynamics in Concept Learning and Compositional Generalization
  Yongyi Yang, Core Francisco Park, Ekdeep Singh Lubana, Maya Okawa, Wei Hu, Hidenori Tanaka
  ICLR 2025

• Abrupt Learning in Transformers: A Case Study on Matrix Completion
  Pulkit Gopalani, Ekdeep Singh Lubana, Wei Hu
  NeurIPS 2024

• Near-Interpolators: Rapid Norm Growth and the Trade-Off between Interpolation and Generalization
  Yutong Wang, Rishi Sonthalia, Wei Hu
  AISTATS 2024

• Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokking
  Kaifeng Lyu*, Jikai Jin*, Zhiyuan Li, Simon S. Du, Jason D. Lee, Wei Hu
  ICLR 2024

• How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representations
  Tianyu Guo, Wei Hu, Song Mei, Huan Wang, Caiming Xiong, Silvio Savarese, Yu Bai
  ICLR 2024

• Benign Overfitting and Grokking in ReLU Networks for XOR Cluster Data
  Zhiwei Xu, Yutong Wang, Spencer Frei, Gal Vardi, Wei Hu
  ICLR 2024

• Bias Amplification Enhances Minority Group Performance
  Gaotang Li*, Jiarui Liu*, Wei Hu
  TMLR 2024

• Going Beyond Linear Mode Connectivity: The Layerwise Linear Feature Connectivity
  Zhanpeng Zhou, Yongyi Yang, Xiaojiang Yang, Junchi Yan, Wei Hu
  NeurIPS 2023

• Are Neurons Actually Collapsed? On the Fine-Grained Structure in Neural Representations
  Yongyi Yang, Jacob Steinhardt, Wei Hu
  ICML 2023

• Implicit Bias in Leaky ReLU Networks Trained on High-Dimensional Data
  Spencer Frei*, Gal Vardi*, Peter L. Bartlett, Nathan Srebro, Wei Hu
  ICLR 2023 (spotlight)

• Representation Alignment in Neural Networks
  Ehsan Imani, Wei Hu, Martha White
  TMLR 2022

• More Than a Toy: Random Matrix Models Predict How Real-World Neural Representations Generalize
  Alexander Wei, Wei Hu, Jacob Steinhardt
  ICML 2022

• A Representation Learning Perspective on the Importance of Train-Validation Splitting in Meta-Learning
  Nikunj Saunshi, Arushi Gupta, Wei Hu
  ICML 2021

• Near-Optimal Linear Regression under Distribution Shift
  Qi Lei, Wei Hu, Jason D. Lee
  ICML 2021

• When is Particle Filtering Efficient for Planning in Partially Observed Linear Dynamical Systems?
  (α-β) Simon S. Du, Wei Hu, Zhiyuan Li, Ruoqi Shen, Zhao Song, Jiajun Wu
  UAI 2021

• Few-Shot Learning via Learning the Representation, Provably
  (α-β) Simon S. Du, Wei Hu, Sham M. Kakade, Jason D. Lee, Qi Lei
  ICLR 2021

• Impact of Representation Learning in Linear Bandits
  Jiaqi Yang, Wei Hu, Jason D. Lee, Simon S. Du
  ICLR 2021

• The Surprising Simplicity of the Early-Time Learning Dynamics of Neural Networks
  Wei Hu, Lechao Xiao, Ben Adlam, Jeffrey Pennington
  NeurIPS 2020 (spotlight)

• Provable Benefit of Orthogonal Initialization in Optimizing Deep Linear Networks
  Wei Hu, Lechao Xiao, Jeffrey Pennington
  ICLR 2020

• Simple and Effective Regularization Methods for Training on Noisily Labeled Data with Generalization Guarantee
  (α-β) Wei Hu, Zhiyuan Li, Dingli Yu
  ICLR 2020

• Enhanced Convolutional Neural Tangent Kernels
  Zhiyuan Li*, Ruosong Wang*, Dingli Yu*, Simon S. Du, Wei Hu, Ruslan Salakhutdinov, Sanjeev Arora
  Manuscript, 2019

• On Exact Computation with an Infinitely Wide Neural Net
  (α-β) Sanjeev Arora, Simon S. Du, Wei Hu, Zhiyuan Li, Ruslan Salakhutdinov, Ruosong Wang
  NeurIPS 2019 (spotlight)

• Implicit Regularization in Deep Matrix Factorization
  (α-β) Sanjeev Arora, Nadav Cohen, Wei Hu, Yuping Luo
  NeurIPS 2019 (spotlight)

• Explaining Landscape Connectivity of Low-cost Solutions for Multilayer Nets
  Rohith Kuditipudi, Xiang Wang, Holden Lee, Yi Zhang, Zhiyuan Li, Wei Hu, Sanjeev Arora, Rong Ge
  NeurIPS 2019

• Nearly Optimal Dynamic k-Means Clustering for High-Dimensional Data
  (α-β) Wei Hu, Zhao Song, Lin F. Yang, Peilin Zhong
  Manuscript, 2019

• Fine-Grained Analysis of Optimization and Generalization for Overparameterized Two-Layer Neural Networks
  (α-β) Sanjeev Arora, Simon S. Du, Wei Hu, Zhiyuan Li, Ruosong Wang
  ICML 2019

• Width Provably Matters in Optimization for Deep Linear Neural Networks
  (α-β) Simon S. Du, Wei Hu
  ICML 2019

• A Convergence Analysis of Gradient Descent for Deep Linear Neural Networks
  (α-β) Sanjeev Arora, Nadav Cohen, Noah Golowich, Wei Hu
  ICLR 2019

• Linear Convergence of the Primal-Dual Gradient Method for Convex-Concave Saddle Point Problems without Strong Convexity
  (α-β) Simon S. Du, Wei Hu
  AISTATS 2019

• Algorithmic Regularization in Learning Deep Homogeneous Models: Layers are Automatically Balanced
  (α-β) Simon S. Du, Wei Hu, Jason D. Lee
  NeurIPS 2018
  (Best paper award at ICML 2018 Workshop on Nonconvex Optimization)

• Online Improper Learning with an Approximation Oracle
  (α-β) Elad Hazan, Wei Hu, Yuanzhi Li, Zhiyuan Li
  NeurIPS 2018

• An Analysis of the t-SNE Algorithm for Data Visualization
  (α-β) Sanjeev Arora, Wei Hu, Pravesh K. Kothari
  COLT 2018

• Linear Convergence of a Frank-Wolfe Type Algorithm over Trace-Norm Balls
  (α-β) Zeyuan Allen-Zhu, Elad Hazan, Wei Hu, Yuanzhi Li
  NIPS 2017 (spotlight)

• Combinatorial Multi-Armed Bandit with General Reward Functions
  (α-β) Wei Chen, Wei Hu, Fu Li, Jian Li, Yu Liu, Pinyan Lu
  NIPS 2016

• New Characterizations in Turnstile Streams with Applications
  (α-β) Yuqing Ai, Wei Hu, Yi Li, David P. Woodruff
  CCC 2016

Blog posts about my work

• Ultra-Wide Deep Nets and Neural Tangent Kernel (NTK)

• Understanding Implicit Regularization in Deep Learning by Analyzing Trajectories of Gradient Descent

• Landscape Connectivity of Low Cost Solutions for Multilayer Nets

• Is Optimization a Sufficient Language for Understanding Deep Learning?

• Understanding Optimization in Deep Learning by Analyzing Trajectories of Gradient Descent

Teaching

University of Michigan

• Fall 2025: EECS 445 – Introduction to Machine Learning

• Winter 2025: CSE 598 – Science of Large Language Models

• Fall 2024: EECS 445 – Introduction to Machine Learning

• Winter 2024: EECS 598 – Machine Learning Theory

• Fall 2023: EECS 445 – Introduction to Machine Learning

• Winter 2023: EECS 598 – Machine Learning Theory

• Fall 2022: EECS 598 – Science of Deep Learning

Princeton University

• Spring 2018: COS 445 – Economics and Computation (teaching assistant)

• Fall 2017: COS 324 – Introduction to Machine Learning (teaching assistant)

Service

• Area chair or equivalent: ICML 2023-2025, NeurIPS 2022-2023, ICLR 2024-2025, CPAL 2024, L4DC 2025

• Conference reviewing: ICML, NeurIPS, ICLR, COLT, FOCS, SODA, AISTATS, ISIT, UAI

• Journal reviewing: Journal of Machine Learning Research, Journal of Artificial Intelligence Research, Machine Learning, Mathematical Programming, Neural Networks, IEEE Signal Processing Letters, Annals of Statistics, Journal of the American Statistical Association

Selected awards

• AAAI New Faculty Highlights, 2024

• Google Research Scholar Award, 2023

• Siebel Scholar, 2021

• Best paper award at ICML Workshop on Modern Trends in Nonconvex Optimization for Machine Learning, 2018

• Gordon Y.S. Wu Fellowship, 2016

• Gold medal (1st place), The 27th Chinese Mathematical Olympiad, 2012