Boya Hou

Research Abstract:

The mature fields of systems and control theory have enabled the success of model-based decision-making. Nonetheless, the techniques involved require analytical modeling of the system, and the complexity of analysis and control grows with that of the model. I am interested in learning compressed representations of dynamical systems from data, in both single-agent and multi-agent contexts. My research focuses on an operator-theoretic approach to capture the action of the system dynamics on suitable spaces of functions--the reproducing kernel Hilbert space--that does not require parametric representations, and thus, makes it appealing from a data-oriented standpoint. Yet the scalabality often suffers with larger datasets. To counter this difficulty, I have explored techniques to reduce redundancy in datasets, and thus, the size/complexity of system models. I studied the sample complexity of such a data-driven approach, illustrating how the relationship between the number of samples and the resulting performance can be controlled. I integrated this approach into reinforcement learning, analysis of global stability, and uncertainty propagation.

Bio:

Boya Hou (she/her) is a PhD candidate with the Department of Electrical and Computer Engineering, University of Illinois, Urbana-Champaign advised by Prof. Subhonmesh Bose. She received the B.Tech. degree from Zhejiang University and the M.Eng. degree from University of Illinois, Urbana-Champaign in 2019. Her research focuses on data-driven learning and decision-making under uncertainty, with applications to electric power grid and sustainable transportation. She is the recipient of the M.A.Pai Scholarship, Mavis Future Faculty Fellowship, AAAI Student Scholarship, and is the second-place winner in the United States Association for Energy Economics (USAEE) Case Competition (2019).