Date: 15 June 2022
Time: 12:45 – 13:15. Hours
Room: RA1501 & online
Speaker: Matus Telgarsky (University of Illinois)
Bio: Matus Telgarsky is an assistant professor at the University of Illinois, Urbana-Champaign, specializing in deep learning theory. He was fortunate to receive a PhD at UCSD under Sanjoy Dasgupta. Other highlights include: co-founding, in 2017, the Midwest ML Symposium (MMLS) with Po-Ling Loh; receiving a 2018 NSF CAREER award; organizing a Simons Institute summer 2019 program on deep learning with Samy Bengio, Aleskander Madry, and Elchanan Mossel.
Title: "The margin approach to the analysis of deep networks”
Abstract:
In this short talk, I'll first pose some approximation-theoretic questions in deep learning, and from there motivate the study of gradient descent, and how it affects approximation-theoretic questions.
Then I will discuss how classical margin theory may be adapted to give convergence guarantees for deep networks, leading to improved analyses not just in standard regimes such as the "neural tangent kernel", but also more general settings.