【Academic Seminar】Momentum-Based Acceleration for Non-convex Stochastic Optimization - Prof. Xuefeng Gao
Topic: Momentum-Based Acceleration for Non-convex Stochastic Optimization
Speaker: Prof. Xuefeng Gao, The Chinese University of Hong Kong
Time and Date: 15:30-16:30, Friday, December 6, 2019
Venue: Boardroom, Dao Yuan Building
Abstract:
We consider stochastic non-convex optimization problems that arise in several applications including machine learning and the stochastic gradient Hamiltonian Monte Carlo (SGHMC) algorithm to solve them. We obtain the first finite-time global convergence guarantees for SGHMC in the context of both empirical and population risk minimization. Our results show that SGHMC can achieve acceleration on a class of non-convex problems compared to overdamped Langevin MCMC approaches such as the stochastic gradient Langevin dynamics. This is a joint work with Mert Gurbuzbalaban from Rutgers University and Lingjiong Zhu from Florida State University.
Biography:
Xuefeng Gao is an Associate Professor at the Department of Systems Engineering and Engineering Management at the Chinese University of Hong Kong. He received his B.S. in Mathematics from Peking University, China in 2008, and his Ph.D. in Operations Research from Georgia Institute of Technology, USA in 2013. His research interests include applied probability, queueing theory, stochastic optimization, and market microstructure.