学术报告第202160期

报告题目：More optimal way to optimize via fractional calculus beyond Nesterov

报 告 人：YangQuan Chen（加利福尼亚大学Merced分校）

报告时间：2021年11月1日 上午10：00--11：00

报告方式：腾讯会议地址：https://meeting.tencent.com/dm/6B593jUBVo2D

会议 ID：379 628 014

主持人：袁烨

报告摘要：

It is now more and more widely accepted that for iterative optimization, it is logical to ask “What is the optimal way to optimize?” Nesterov accelerated gradient method (AGM) is introduced first and then Michael I Jordan’s ICM18 talk is briefly reviewed. We reexamine the iteration-axis dynamics from control system point of view using z-transform and show that Nesterov scheme is a second order (lead-lag) dynamic system along iteration axis. In continuum limit, continuous-time system theory can be used for designing new iteration schemes. As an example, we introduce a continuous-time fractional order filter and then discretize it to form a new iterative scheme, resulting a potentially more optimal way to optimize. Stochastic AGM will also be discussed briefly with an indication of a potential benefit of using fractional order stochasticity in achieving more optimal way to optimize. It is hoped that this talk will open new research opportunities in machine learning in general and deep artificial neural network learning in particular.

报告人简介：

YangQuan Chen earned his Ph.D. from Nanyang Technological University, Singapore, in 1998. He had been a faculty of Electrical Engineering at Utah State University (USU) from 2000-12. He joined the School of Engineering, University of California, Merced (UCM) in summer 2012 teaching “Mechatronics”, “Engineering Service Learning” and “Unmanned Aerial Systems” for undergraduates; “Fractional Order Mechanics”, “Linear Multivariable Control”, “Nonlinear Controls” and “Advanced Controls: Optimality and Robustness” for graduates. His research interests include mechatronics for sustainability, cognitive process control (smart control engineering enabled by digital twins), small multi-UAV based cooperative multi-spectral “personal remote sensing”, applied fractional calculus in controls, modeling and complex signal processing; distributed measurement and control of distributed parameter systems with mobile actuator and sensor networks. He received Research of the Year awards from both USU (2012) and UCM (2020). He was listed in Highly Cited Researchers by Clarivate Analytics in 2018-2021. His lab website is http://mechatronics.ucmerced.edu/Google scholar page: https://scholar.google.com/citations?user=RDEIRbcAAAAJ&hl=en