MEchanics GAthering –MEGA- Seminar: Bayesian Optimization for Performance Auto-Tuning: Constrained, Grey-Box, and Multi-Agent Algorithms
Abstract
Performance tuning is ubiquitous in engineering, where a set of parameters are tuned to optimize the performance of a system subject to a set of constraints to be satisfied. It is of great practical interest to automate this process by algorithms to boost the tuning efficiency. However, the mapping from the parameters to the performance/constraints are typically black-box functions and each evaluation of these functions can be very expensive. Bayesian optimization (BO), as a sample-efficient black-box optimization method, has recently found successful applications in many different performance tuning tasks. However, conventional BO algorithms can not manage the constraint violation well and suffer from the curse of dimensionality. In this talk, I will present our recent works on constrained, grey-box and multi-agent Bayesian optimization algorithms, in an attempt to address the violation management and scalability issues.
Biography
Wenjie Xu is a third-year Ph.D. student in Electrical Engineering with the Automatic Control Laboratory, EPFL, and a guest Ph.D. student with the Urban Energy Systems Laboratory at Empa. He is jointly supervised by Prof. Colin Jones at EPFL and Dr. Bratislav Svetozarevic from Empa. He received MPhil degree from The Chinese University of Hong Kong in 2020 and B.E. degree from Tsinghua University in 2018. His research interests include data-driven optimization, controller performance tuning, intelligent transportation, and smart energy systems, with more than ten papers published in journals (such as T-ITS, T-NNLS, etc.) and conferences (such as ICML, MobiHoc, CDC, etc.). He received the ASME Energy Systems Technical Committee Best Paper Award at the 2022 American Control Conference.
Performance tuning is ubiquitous in engineering, where a set of parameters are tuned to optimize the performance of a system subject to a set of constraints to be satisfied. It is of great practical interest to automate this process by algorithms to boost the tuning efficiency. However, the mapping from the parameters to the performance/constraints are typically black-box functions and each evaluation of these functions can be very expensive. Bayesian optimization (BO), as a sample-efficient black-box optimization method, has recently found successful applications in many different performance tuning tasks. However, conventional BO algorithms can not manage the constraint violation well and suffer from the curse of dimensionality. In this talk, I will present our recent works on constrained, grey-box and multi-agent Bayesian optimization algorithms, in an attempt to address the violation management and scalability issues.
Biography
Wenjie Xu is a third-year Ph.D. student in Electrical Engineering with the Automatic Control Laboratory, EPFL, and a guest Ph.D. student with the Urban Energy Systems Laboratory at Empa. He is jointly supervised by Prof. Colin Jones at EPFL and Dr. Bratislav Svetozarevic from Empa. He received MPhil degree from The Chinese University of Hong Kong in 2020 and B.E. degree from Tsinghua University in 2018. His research interests include data-driven optimization, controller performance tuning, intelligent transportation, and smart energy systems, with more than ten papers published in journals (such as T-ITS, T-NNLS, etc.) and conferences (such as ICML, MobiHoc, CDC, etc.). He received the ASME Energy Systems Technical Committee Best Paper Award at the 2022 American Control Conference.
Practical information
- General public
- Free
Organizer
- MEGA.Seminar Organizing Committee