Colloquiums and Conferences

Stochastic approximation of functions by neural networks in kernel-regression formations

Time: 2025-01-21 20:59:00

Title: Stochastic approximation of functions by neural networks in kernel-regression formations
Speaker: Prof. Xingping Sun (Invited by Prof. Qi Ye)
Affiliation: Missouri State University, USA
Time: January 21, 10:30–11:30
Venue: Lecture Hall, Room 111, West Building, School of Mathematical Sciences

Speaker Biography:
    Xingping Sun is a Distinguished Professor at Missouri State University, where he has worked for many years on approximation theory, statistical machine learning and radial basis functions. He received his Ph.D. in mathematics from the University of Texas at Austin in 1990, after earning bachelor's and master's degrees at Henan Normal University and Northwest University in China. He joined Missouri State University (then Southwest Missouri State University) in 1990, was promoted to full professor in 2000 and appointed Distinguished Professor in 2020, and has held visiting professorships at Texas A&M University, Vanderbilt University and the University of Leicester, among others. He has published extensively in leading journals such as Mathematics of Computation, SIAM Journal on Numerical Analysis and Journal of Approximation Theory, and has led numerous US National Science Foundation (NSF) and international collaborative research projects. He also serves on several editorial boards and conference committees. He was Associate Dean of the College of Natural and Applied Sciences at Missouri State University (2012–2015) and has received multiple teaching and research awards.

Abstract:
    Universal approximation theorems in the field of machine learning state that neural networks with some common mathematical structures are employable for the purpose of approximating continuous functions. However, most of these theorems are non-constructive in nature, and therefore have left ample rooms for improvement. In this talk, we propose and study a constructive method for stochastic approximation of functions by neural networks, which features kernel-regression formations.