연구성과
Adaptive Bayesian Optimization for Fast Exploration Under Safety Constraints
- 작성자관리자
- 작성일2023.09.07
- 조회수104
Guk Han, Jeongoh Jeong, Jong-Hwan Kim
상세내용
BibTeX
- 저자 : Guk Han, Jeongoh Jeong, Jong-Hwan Kim
- 논문명 : Adaptive Bayesian Optimization for Fast Exploration Under Safety Constraints
- 학술지명 : IEEE Access
- 발간월 : April
- SCI-E 여부 : SCI-E
- ISSN : ISSN
- DOI : https://doi.org/10.7746/jkros.2022.17.1.016
- 초록 : The industrial field faces the problem of process optimization by finding the factors affecting the yield of the process and controlling them appropriately. However, due to limited resources such as time and money, optimization is performed using a low evThe industrial field faces the problem of process optimization by finding the factors affectingthe yield of the process and controlling them appropriately. However, due to limited resources such as timeand money, optimization is performed using a low evaluation budget. In addition, for process stability, thelower limit of the yield is set so that the yield must be maintained above this limit during optimization.Bayesian Optimization (BO) can be an effective solution in acquiring optimal samples that satisfy a safetyconstraint given a low evaluation budget. However, many existing BO algorithms have some limitations suchas significant performance degradation due to model misspecification, and high computational load.Thus,we propose a practical safe BO algorithm, A-SafeBO, that effectively reduces performance degradationdue to model misspecification using only a limited evaluation budget. Additionally, our algorithm performscomputations for a large number of observations and high-dimensional input spaces by using EnsembleGaussian Processes and Safe Particle Swarm Optimization. Here, we also propose a new acquisition functionthat leads to a wider exploration even under the constraint of safety. This will help deviate from the localoptimum and achieve a better recommendation. Our algorithm empirically guarantees convergence andperformance through evaluations on several synthetic benchmarks and a real-world optimization problem