Exploiting Diversity for Online Learning with Adaptivity and Robustness

Qin Lu, Assistant Professor, School of Electrical and Computer Engineering, University of Georgia

1400 Wegmans Hall

Abstract: Function approximation is a key module underlying all machine learning algorithms. In many safety-critical applications where uncertainty quantification of the point function estimate is called for, one would resort to the so-termed Gaussian process (GP), a well-established Bayesian nonparametric function learning framework that provides well-calibrated uncertainty values. Building on the GP paradigm, this talk will cover two related and complementary online learning tasks, namely, supervised prediction and Bayesian (bandit) optimization. Specifically, I will show how to ensemble (E) a diversity of GP learners for adaptivity in these two online tasks. Besides, I will present associated regret-based theoretical results to demonstrate the robustness of the EGP-based approaches. Lastly, empirical results on real datasets will be presented to validate the merits of the proposed methods relative to existing baselines. 

Bio: Qin Lu is an assistant professor with the School of Electrical and Computer Engineering at the University of Georgia. Previously, she worked as a postdoctoral research associate at the University of Minnesota, Twin Cities. She received her B.S. and Ph.D. degrees in electrical engineering from the University of Electronic Science and Technology of China and the University of Connecticut (UConn) in 2013 and 2018, respectively. Her research interests span the areas of signal processing, machine learning, data science, and communications, with special focus on Gaussian processes, Bayesian optimization, spatio-temporal inference over graphs, and data association for multi-object tracking. She received the National Scholarship from China twice. She was awarded Summer Fellowship and Doctoral Dissertation Fellowship at UConn.  She was also a recipient of the Women of Innovation Award in Collegian Innovation and Leadership by Connecticut Technology Council in 2018.

Refreshments will be provided.