TY - GEN
T1 - Using Gaussian Processes to Automate Probabilistic Branch Bound for Global Optimization
AU - Pedrielli, Giulia
AU - Huang, Hao
AU - Zabinsky, Zelda B.
N1 - Funding Information:
*This work was supported by NSF CMMI#, NSF CISE # 1Giulia Pedrielli is with School of Computing Informatics and Decision Systems Engineering, Arizona State University, 699 S Mill Ave, Tempe, AZ 85258, USA giulia.pedrielli@asu.edu 2Hao Huang is with College of Engineering, Yuan Ze University, TAIWAN haohuang@saturn.yzu.edu.tw 3Zelda B. Zabinsky is with the Department of Industrial and Systems Engineering, University of Washington, Seattle, WA 98195-2650, USA zelda@uw.edu
Funding Information:
ACKNOWLEDGMENT The research in this paper has been partially supported by the grant NSF #2000792, and NSF #1829238.
Publisher Copyright:
© 2021 IEEE.
PY - 2021/8/23
Y1 - 2021/8/23
N2 - Manufacturing, aerospace, energy and several other industries have witnessed a steep growth of increasingly complex, information rich, devices and systems of devices requiring simulation-based approaches. In fact, most modern systems have such complex behavior that their performance can only be evaluated through, usually computationally expensive, simulations. In such settings, it is of paramount importance to provide solutions with quality guarantees. In this manuscript, we focus on algorithms capable of identifying a level set of solutions in proximity of the global optimum, and specifically on the Probabilistic Branch and Bound (PBnB) method. We propose a new way to automate branching decisions by coupling this method with Gaussian process (GP) estimation. The result is PBnB-GP, where, at each iteration a collection of GPs is used to decide how to branch the input space. PBnB-GP not only returns an estimate of the regions with near-optimal reward (using the power of PBnB), but also a 'collection of Gaussian processes' that can produce point estimations for any location in the input space, thus harnessing the power of model-based black-box optimization. We present PBnB-GP for the first time together with preliminary numerical results.
AB - Manufacturing, aerospace, energy and several other industries have witnessed a steep growth of increasingly complex, information rich, devices and systems of devices requiring simulation-based approaches. In fact, most modern systems have such complex behavior that their performance can only be evaluated through, usually computationally expensive, simulations. In such settings, it is of paramount importance to provide solutions with quality guarantees. In this manuscript, we focus on algorithms capable of identifying a level set of solutions in proximity of the global optimum, and specifically on the Probabilistic Branch and Bound (PBnB) method. We propose a new way to automate branching decisions by coupling this method with Gaussian process (GP) estimation. The result is PBnB-GP, where, at each iteration a collection of GPs is used to decide how to branch the input space. PBnB-GP not only returns an estimate of the regions with near-optimal reward (using the power of PBnB), but also a 'collection of Gaussian processes' that can produce point estimations for any location in the input space, thus harnessing the power of model-based black-box optimization. We present PBnB-GP for the first time together with preliminary numerical results.
UR - http://www.scopus.com/inward/record.url?scp=85117030902&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85117030902&partnerID=8YFLogxK
U2 - 10.1109/CASE49439.2021.9551592
DO - 10.1109/CASE49439.2021.9551592
M3 - Conference contribution
AN - SCOPUS:85117030902
T3 - IEEE International Conference on Automation Science and Engineering
SP - 2276
EP - 2281
BT - 2021 IEEE 17th International Conference on Automation Science and Engineering, CASE 2021
PB - IEEE Computer Society
T2 - 17th IEEE International Conference on Automation Science and Engineering, CASE 2021
Y2 - 23 August 2021 through 27 August 2021
ER -