Abstract

With the growing amount ofsensing data and increasing level of system complexity, there are emergingopportunities and challenges in designing reliable operation schemes forvarious tasks in cyber-physical systems such as robotics manipulation, buildingenergy management and optimal power flow. This calls for a rethinking ofcontrol and optimization theory from the standpoint of data-driven approach.

 In the first part of the talk,I will focus on learning to solve optimization considering the datauncertainties, where the rich information of the underlying optimizationproblem along with the convex optimization theory provide design modules for anefficient learner to output the solutions satisfying all engineeringconstraints.

In the second part, I willtouch on the decision-making problem where a model is not known a priori, andinvestigate how specifically designed neural networks can provide optimalcontrol solutions. This result represents the design principles for learning anaccurate and computationally tractable controller. Together, these exampleshighlight the important role of learning and closed-loop control in the designof sustainable and reliable information and physical systems.

Time

2月2日（星期二）09:30-11:30

Speaker

Yize Chen isa Ph.D. candidate from the Department of Electrical and Computer Engineering atUniversity of Washington, and he received his B.S. with honors from Chu KochenCollege at Zhejiang University in 2016. His research lies in control, learningand optimization of cyber-physical systems with performance guarantees. He isthe recipient of several awards, including the 2019 ACM e-Energy Best PaperRunnerUp. He held internship and research positions at Harvard Medical School,Los Alamos National Laboratory, and Microsoft Research. His research issupported by Keith and Nancy Rattie Fellowship and University of WashingtonClean Energy Institute Fellowship. Homepage: https://blogs.uw.edu/yizechen/

Venue

Zoom会议室号码：99781594966

密码：431692