Mathematical and Numerical Understanding of Neural Networks: From Representation to Learning Dynamics and From Shallow to Deep
Professor Hongkai Zhao
, Duke University
Abstract:
In this talk, Dr. Hongkai Zhao will present both mathematical and numerical analysis as well as experiments to study a few basic computational issues in using neural networks to approximate functions: (1) frequency bias in terms of representation and learning, (2) stability and accuracy, and (3) structured and balanced approximation using multi-component and multi-layer neural networks (MMNN).
Speaker’s Bio:
Hongkai Zhao is a Ruth F. DeVarney Distinguished Professor of Mathematics at Duke University. He was formerly the Chancellor's Professor in the Department of Mathematics at the University of California, Irvine. He is known for his work in scientific computing, imaging and numerical analysis, such as the fast sweeping method for Hamilton-Jacobi equation and numerical methods for moving interface problems.
