Skip to content
Title: Over-parameterization in Deep Learning: Kernel Regime and Beyond
Speaker: Dr. Difan Zou, Assistant Professor, IDS & Department of Computer Science, HKU
Date: Apr 12, 2023
Time: 3:30pm – 4:30pm

Mode: Hybrid. Seats for on-site participants are limited. A confirmation email will be sent to participants who have successfully registered.

Seminar recording:


In recent years, deep learning has revolutionized the field of artificial intelligence, achieving remarkable success in a variety of applications. However, the high-dimensional and nonconvex nature of deep neural networks has made it challenging to understand their behavior and performance. Over-parameterization, which refers to the phenomenon where neural networks have more parameters than necessary to fit the training data, has become a key concept in the study of deep learning. In this talk, the speaker will explore the concept of over-parameterization and introduce a series of recent works that fall in the so-called “kernel regime where the neural network behaves like a kernel method. Furthermore,I will discuss the advantages and limitations of these kernel-based analyses, and introduce several remarkable attempts beyond the kernel regime. Specifically, the speaker will discuss how over-parameterization can affect generalization, optimization, and sample complexity. Overall, this talk aims to provide a comprehensive overview of over-parameterization in deep learning, and to highlight key questions as well as further research directions in this exciting and rapidly evolving area.


Dr. Difan Zou
Assistant Professor @ HKU IDS & Dept of CS
Dr. Difan Zou is an Assistant Professor in the HKU IDS and Department of Computer Science at the University of Hong Kong. He received his Ph.D. in Computer Science, University of California, Los Angeles (UCLA). He received a B. S degree in Applied Physics, from School of Gifted Young, USTC and a M. S degree in Electrical Engineering from USTC. He has published multiple papers on top-tier machine learning conferences including ICML, NeurIPS, ICLR, COLT, etc. He is a recipient of Bloomberg Data Science Ph.D. fellowship.
His research interests are broadly in machine learning, optimization, and learning structured data (e.g., time-series or graph data), with a focus on theoretical understanding of the optimization and generalization in deep learning problems.

For information, please contact: