PeopleofSuperintelligence

Researcher

Jingfeng Wu

Also known as 吴京风

Jingfeng Wu is an assistant professor of statistics at UC Berkeley whose work covers deep learning theory, optimization and statistical learning. He earned a mathematics bachelor’s degree at Peking University in 2016, a master’s in applied mathematics there in 2019, and a computer science PhD at Johns Hopkins University in 2023. From 2023 to 2026, he was a postdoctoral fellow at Berkeley’s Simons Institute, hosted by Peter Bartlett and Bin Yu. His papers examined when stochastic gradient descent generalizes in linear regression in 2021 and how large training steps can speed up gradient descent for logistic regression in 2024. In December 2025, he presented a NeurIPS tutorial on training instability with Yu-Xiang Wang and Maryam Fazel.

Works
38
Citations
486
h-index
8

Publication numbers from OpenAlex, as of 28 October 2026.

Timeline

  1. 2016–2019

    Mathematics degrees at Peking

    He earns a BS in mathematics at Peking University in 2016 and an MS in applied mathematics there in 2019.

  2. 2021

    Benign overfitting with SGD

    With Difan Zou, Vladimir Braverman, Quanquan Gu and Sham Kakade, he coauthors Benign Overfitting of Constant-Stepsize SGD for Linear Regression in 2021. The paper gives bounds on prediction error that describe when the method can generalize in overparameterized linear regression.

  3. 2023

    Computer science PhD at Hopkins

    In 2023, he completes a PhD in computer science at Johns Hopkins University, advised by Vladimir Braverman.

  4. 2023–2026

    Postdoctoral fellow at Simons

    From 2023 to 2026, he works as a postdoctoral fellow at UC Berkeley’s Simons Institute for the Theory of Computing, hosted by Peter Bartlett and Bin Yu. He takes part in the NSF/Simons Collaboration on the Theoretical Foundations of Deep Learning.

  5. 2024

    Large steps in gradient descent

    In 2024, he coauthors Large Stepsize Gradient Descent for Logistic Loss with Peter Bartlett, Matus Telgarsky and Bin Yu. For logistic regression on linearly separable data, the paper shows that large constant steps can speed up gradient descent after an initial phase of oscillating loss.

  6. February 2025

    Organizing deep learning theory

    In February 2025, he organizes the Deep Learning Theory workshop at the Simons Institute with Peter Bartlett and Soufiane Hayou. The workshop brings together researchers to develop mathematical, statistical and algorithmic tools for understanding deep learning.

  7. December 2025

    Teaching about training instability

    In December 2025, he presents the NeurIPS tutorial Theoretical Insights on Training Instability in Deep Learning with Yu-Xiang Wang and Maryam Fazel. He teaches the section on how large step sizes accelerate optimization.

  8. 2026

    Comparing regression prediction risks

    His paper Risk Comparisons in Linear Regression: Implicit Regularization Dominates Explicit Regularization is accepted for presentation at COLT 2026. Written with Peter Bartlett, Sham Kakade, Jason Lee and Bin Yu, the work compares the prediction risks of gradient descent, ridge regression and online stochastic gradient descent for individual linear regression problems.

  9. Since 2026

    Assistant professor at Berkeley

    In 2026, he becomes an assistant professor in UC Berkeley’s Department of Statistics. His research covers machine learning, deep learning theory, optimization and statistical learning theory.

Career

  1. Since 2026University of California, BerkeleyAssistant Professor
  2. 2023–2026University of California, BerkeleyPostdoctoral Fellow at the Simons Institute for the Theory of Computing

Education

  1. 2023Johns Hopkins UniversityPhD in computer science
  2. 2019Peking UniversityMS in applied mathematics
  3. 2016Peking UniversityBS in mathematics

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Jingfeng Wu yet. Send us a free-licensed photo.

  1. CSE Faculty Candidate Seminar - Jingfeng Wu | School of Interactive Computing ic.gatech.edu. Checked 28 October 2026.
  2. Benign Overfitting of Constant-Stepsize SGD for Linear Regression arxiv.org. Checked 28 October 2026.
  3. Jingfeng Wu uuujf.github.io. Checked 28 October 2026.
  4. Jingfeng Wu simons.berkeley.edu. Checked 28 October 2026.
  5. Large Stepsize Gradient Descent for Logistic Loss: Non-Monotonicity of the Loss Improves Optimization Efficiency arxiv.org. Checked 28 October 2026.
  6. Deep Learning Theory simons.berkeley.edu. Checked 28 October 2026.
  7. NeurIPS Tutorial Theoretical Insights on Training Instability in Deep Learning neurips.cc. Checked 28 October 2026.
  8. Risk Comparisons in Linear Regression: Implicit Regularization Dominates Explicit Regularization arxiv.org. Checked 28 October 2026.
  9. Jingfeng Wu | Department of Statistics statistics.berkeley.edu. Checked 28 October 2026.