who: Zi Wang

when: 10/31 (Thursday) 11:45a – 1:15p

where: Computer Science Building Rm 150

food: Athena’s Pizza
generous sponsor:  ORACLE LABS

Bayesian Optimization for Global Optimization of Expensive Black-box Functions

Abstract:

Many problems in areas ranging from finance and product design to engineering in general all boil down to the problem of optimizing expensive black-box functions. Bayesian optimization uses probabilistic methods to address this problem with assumptions usually expressed by a Gaussian process prior. Motivated by real-world applications in high-dimensional parameter-tuning problems for complex machine learning algorithms and expensive active learning problems in robotics, we study the theoretical understandings of Bayesian optimization, connections among existing methods, and develop efficient and provably correct Bayesian optimization methods for these applications. In this talk, I will give an in-depth tour of our study of Bayesian optimization on how to design a better data acquisition strategy, how to scale up the method to higher-dimensional and larger-scale data and how to analyze the sample complexity without assuming the full knowledge of the prior. Finally, I will also briefly show how we utilized some of these ideas to tackle problems in robot learning and planning for complex long-horizon problems.

Bio:

Zi Wang is a Ph.D. candidate at MIT Computer Science and Artificial Intelligence Laboratory advised by Prof. Leslie Pack Kaelbling and Prof. Tomas Lozano-Perez. Her PhD research focuses on tackling problems related to robot learning, active learning for planning and Bayesian optimization. She received her M.S. degree in Electrical Engineering and Computer Science from MIT in Feb 2016 and B.Eng. degree in Computer Science and Technology from Tsinghua University in Jul 2014. Zi is a recipient of MIT Graduate Women of Excellence Award, Rising Star in EECS and Google Anita Borg Scholarship. While at MIT, she served as co-president of Graduate Women in Course 6 (EECS), co-organizer of the first Machine Learning Across MIT Retreat and research mentor for several undergraduate and MEng students.