Learning Overparametrized Neural Networks and Statistical Models

2021-06-17 14:00-15:30 Science Building No.1, 1114

Abstract: Modern machine learning has constantly presented puzzling empirical properties and surprised the classical statistical theory. Learning with overparametrized models is becoming a norm in data-analytic applications, and the tension of memorization rarely bothers practitioners. In this talk, I will discuss the training of overparametrized neural networks from both the neural tangent kernel and the mean-field perspectives, which guarantees the global convergence property despite the non-convexity of the optimization landscape. I will also discuss more interesting phenomena in a series of overparametrized statistical questions.


Bio: Pengkun Yang is an assistant professor in the Center for Statistical Science at Tsinghua University. Prior to joining Tsinghua, he was a Postdoctoral Research Associate in the Department of Electrical Engineering at Princeton University. He received a Ph.D. degree (2018) and a master’s degree (2016) from the Department of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign, and a B.E. degree (2013) from the Department of Electronic Engineering at Tsinghua University. His research interests include statistical inference, learning, optimization, and systems. He is a recipient of Thomas M. Cover Dissertation Award in 2020, and a recipient of Jack Keil Wolf ISIT Student Paper Award at the 2015 IEEE International Symposium on Information Theory (semi-plenary talk).