Decision and Control Lecture Series || Decision-Making and Learning Under Dynamic Information: Roy Dong UIUC

Abstract:

In many modern settings, one must decide when to take an action with payoffs and uncertainties evolving over time. For example, an agent may have to act on a timescale much faster than its estimation and perception algorithms. Alternatively, a consumer may be learning more about her valuation of a good while a personalized pricing algorithm works to extract maximal revenue from her. Furthermore, as one makes data-driven decisions, the closed-loop effects of these decisions may affect the underlying data generating processes, e.g., a strategic data source may change its reported data to improve their assigned label. First, we'll present our recent work on duality approaches for the discrete-time optimal stopping problem, which provides new computational insights, as well a method to analyze the effect of changes in payoff and information on the optimal stopping strategies. Next, we analyze the decision-dependent distribution shift through the lens of performative prediction and show how tools from perturbation analysis in control theory can be applied to this problem space. These works are part of our broader research goal of understanding the dynamic role of information and learning in closed-loop settings.

Biography:

Roy Dong is a Research Assistant Professor in the Electrical and Computer Engineering department at the University of Illinois at Urbana-Champaign. He received a BS Honors in Computer Engineering and a BS Honors in Economics from Michigan State University in 2010. He received a PhD in Electrical Engineering and Computer Sciences at the University of California, Berkeley in 2017, where he was funded in part by the NSF Graduate Research Fellowship. From 2017 to 2018, he was a postdoctoral researcher in the Berkeley Energy & Climate Institute and a visiting lecturer in the Industrial Engineering and Operations Research department at UC Berkeley. His research uses tools from control theory, economics, statistics, and optimization to understand the closed-loop effects of machine learning, with applications in cyber-physical systems such as the smart grid, modern transportation networks, and autonomous vehicles.