Mo Zhou 周墨 Postdoc Institute for Foundations of Data Science (IFDS) University of Washington Email: mozhou17 at cs dot washington dot edu Google scholar

I am a postdoc at Institute for Foundations of Data Science (IFDS) at University of Washington, where I'm working with Simon Du and Maryam Fazel. Previously, I received my Ph.D. in Computer Science from Duke University, where I was fortunate to be advised by Rong Ge. Before coming to Duke, I received B.S. in Statistics from Peking University.

In summer 2023, I worked with Prof. Tengyu Ma at Stanford. In summer 2022, I was an applied science intern at AWS AI. In summer 2018, I was an intern in Industrial and Systems Engineering (ISyE), Georgia Tech, working with Prof. Tuo Zhao.

My research interests are in optimization and theoretical machine learning. Recently, I am particularly interested in deep learning theory.

Publications and Preprints

* denotes equal contribution, (α-β order) denotes alphabetical ordering

• How Does Gradient Descent Learn Features -- A Local Analysis for Regularized Two-Layer Neural Networks
  Mo Zhou, Rong Ge.
  Conference on Neural Information Processing Systems (NeurIPS), 2024.
  Short version appeared at NeurIPS Mathematics of Modern Machine Learning (M3L) workshop, 2023.


• Implicit Regularization Leads to Benign Overfitting for Sparse Linear Regression
  Mo Zhou, Rong Ge.
  International Conference on Machine Learning (ICML), 2023.


• Understanding Edge-of-Stability Training Dynamics with a Minimalist Example
  Xingyu Zhu*, Zixuan Wang*, Xiang Wang, Mo Zhou, Rong Ge.
  International Conference on Learning Representations (ICLR), 2023


• Depth-Separation with Multilayer Mean-Field Networks.
  Yunwei Ren, Mo Zhou, Rong Ge.
  International Conference on Learning Representations (ICLR), 2023. Notable-top-25%.


• Plateau in Monotonic Linear Interpolation–A “Biased” View of Loss Landscape for Deep Networks
  Xiang Wang, Annie N Wang, Mo Zhou, Rong Ge.
  International Conference on Learning Representations (ICLR), 2023


• Understanding The Robustness of Self-supervised Learning Through Topic Modeling
  Zeping Luo*, Shiyou Wu*, Cindy Weng*, Mo Zhou, Rong Ge
  International Conference on Learning Representations (ICLR), 2023


• Understanding Deflation Process in Over-parametrized Tensor Decomposition
  (α-β order) Rong Ge*, Yunwei Ren*, Xiang Wang*, Mo Zhou*
  Conference on Neural Information Processing Systems (NeurIPS), 2021.


• A Local Convergence Theory for Mildly Over-Parameterized Two-Layer Neural Network
  Mo Zhou, Rong Ge, Chi Jin
  Conference on Learning Theory (COLT), 2021.


• Towards Understanding the Importance of Shortcut Connections in Residual Networks
  Tianyi Liu*, Minshuo Chen*, Mo Zhou, Simon S. Du, Enlu Zhou, Tuo Zhao
  Conference on Neural Information Processing Systems (NeurIPS), 2019.


• Towards Understanding the Importance of Noise in Training Neural Networks
  Mo Zhou*, Tianyi Liu*, Yan Li, Dachao Lin, Enlu Zhou, Tuo Zhao
  International Conference on Machine Learning (ICML), 2019. Long Talk.

Presentations

• A Local Convergence Theory for Mildly Over-Parameterized Two-Layer Neural Network
  COLT 2021, Aug. 2021
  Theory of Overparameterized Machine Learning (TOPML) 2021, Apr. 2021
  Duke Deep Learning Reading Group, Apr. 2021
  THEORINET Journal Club/MODL Reading Group, Feb. 2021
  

Teaching

• CPS590.04 Machine Learning Algorithms, 2021 Spring. TA


• CPS330 Design and Analysis of Algorithms, 2020 Fall. TA


• CPS330 Design and Analysis of Algorithms, 2020 Spring. TA

Services

• Reviewer for ICML, ICLR, NeurIPS, JMLR, Mathematical Programming, STOC.

Education

• Duke University, 2019 - 2024
  Ph.D. in Computer Science


• Peking University, 2015 - 2019
  B.S. in Statistics

Last update: Oct. 2024