Ziwei Ji

I am a research scientist at Google Research, New York City. Previously, I received my PhD degree from University of Illinois Urbana-Champaign, advised by Matus Telgarsky. I obtained my bachelor's degree in Computer Science from the ACM class at Shanghai Jiao Tong University. I am interested in machine learning and optimization, particularly in deep learning theory.

Email: ziweiji (at) google (dot) com

CV. Google Scholar.

Publications

Conference publications

• Reproducibility in Optimization: Theoretical Framework and Limits
  Kwangjun Ahn, Prateek Jain, Ziwei Ji, Satyen Kale, Praneeth Netrapalli, Gil I. Shamir.
  NeurIPS 2022, Oral.

• Agnostic Learnability of Halfspaces via Logistic Loss
  Ziwei Ji, Kwangjun Ahn, Pranjal Awasthi, Satyen Kale, Stefani Karp.
  ICML 2022, Long Presentation.

• Actor-critic is implicitly biased towards high entropy optimal policies
  Yuzheng Hu, Ziwei Ji, Matus Telgarsky.
  ICLR 2022.

• Early-stopped neural networks are consistent
  Ziwei Ji, Justin D. Li, Matus Telgarsky.
  NeurIPS 2021, Spotlight.

• Fast Margin Maximization via Dual Acceleration
  Ziwei Ji, Nathan Srebro, Matus Telgarsky.
  ICML 2021.

• Generalization bounds via distillation
  Daniel Hsu, Ziwei Ji, Matus Telgarsky, Lan Wang.
  ICLR 2021, Spotlight.

• Characterizing the implicit bias via a primal-dual analysis
  Ziwei Ji, Matus Telgarsky.
  ALT 2021.

• Directional convergence and alignment in deep learning
  Ziwei Ji, Matus Telgarsky.
  NeurIPS 2020, Spotlight.

• Gradient descent follows the regularization path for general losses
  Ziwei Ji, Miroslav Dudík, Robert E. Schapire, Matus Telgarsky.
  COLT 2020.

• Polylogarithmic width suffices for gradient descent to achieve arbitrarily small test error with shallow ReLU networks
  Ziwei Ji, Matus Telgarsky.
  ICLR 2020.

• Neural tangent kernels, transportation mappings, and universal approximation
  Ziwei Ji, Matus Telgarsky, Ruicheng Xian.
  ICLR 2020.

• Risk and parameter convergence of logistic regression
  Ziwei Ji, Matus Telgarsky.
  In COLT 2019 under the name “The implicit bias of gradient descent on nonseparable data”.

• Gradient descent aligns the layers of deep linear networks
  Ziwei Ji, Matus Telgarsky.
  ICLR 2019.

• Social Welfare and Profit Maximization from Revealed Preferences
  Ziwei Ji, Ruta Mehta, Matus Telgarsky.
  WINE 2018.

Talks

• Agnostic Learnability of Halfspaces via Logistic Loss
  Seminar in Statistical Learning, Dynamical Systems and Probability, AI Institute in Leipzig, October 2022.

• The dual of the margin: improved analyses and rates for gradient descent’s implicit bias
  One World Seminar Series on the Mathematics of Machine Learning, December 2020.

• Polylogarithmic width suffices for gradient descent to achieve arbitrarily small test error with shallow ReLU networks
  Illinois Institute for Data Science and Dynamical Systems (IDS2) Seminar Series, April 2020;
  14th Annual Machine Learning Symposium, The New York Academy of Sciences, March 2020;
  15th CSL Student Conference, February 2020.

Teaching

• Teaching assistant for UIUC CS 598 Deep Learning Theory (Fall 2020, Fall 2021).

• Teaching assistant for UIUC CS 446 Machine Learning (Spring 2019).

Service

Reviewer for NeurIPS, ICLR, COLT, ICML, EC, ITCS, IEEE Transactions on Information Theory.