In this lecture, Peter Frazier of Cornell University will discuss Bayesian optimization and its applications.
Grey-box Bayesian Optimization
Presented by Peter Frazier, Cornell University
Friday, November 15, 2019
Brickyard (BYENG) 510, Tempe campus [map]
Bayesian optimization is a powerful tool for optimizing time-consuming-to-evaluate nonconvex derivative-free objective functions. While BayesOpt has historically been deployed as a blackbox optimizer, recent advances show considerable gains by “peeking inside the box”. For example, when tuning hyperparameters in deep neural networks to minimize validation error, state-of-the-art BayesOpt tuning methods leverage the ability to stop training early, restart previously paused training, perform training and testing on a strict subset of the available data, and warm-start from previously tuned network architectures. New “grey box” Bayesian optimization methods that selectively exploit problem structure to deliver state-of-the-art performance will be described. Afterwards applications of these methods to tuning deep neural networks, inverse reinforcement learning and calibrating physics-based simulators to observational data will be described.
About the speaker
Peter Frazier is an Associate Professor in the School of Operations Research and Information Engineering at Cornell University and a Staff Data Scientist at Uber. His academic research is on Bayesian optimization, multi-armed bandits and reinforcement learning with applications in e-commerce, the sharing economy and materials design. At Uber, he managed UberPool’s data science group, designed the route-based pricing portion of Uber’s pricing system, and currently works on features that increase drivers’ flexibility and control. He is the recipient of an AFOSR Young Investigator Award, an NSF CAREER Award, and best paper awards from the ACM Conference on Economics and Computation, Winter Simulation Conference, and the INFORMS Computing Society.