Robust prior-biased acquisition function for human-in-the-loop Bayesian optimization

作     者：Guay-Hottin, Rose Kardassevitch, Lison Pham, Hugo Lajoie, Guillaume Bonizzato, Marco 

作者机构：Polytech Montreal Elect Engn Dept Montreal PQ H3T 1J4 Canada MILA Quebec Inst Montreal PQ H2S 3H1 Canada Univ Montreal Neurosci Dept Montreal PQ H3T 1J4 Canada Univ Montreal Ctr interdisciplinaire Rech Cerveau & Apprentissag Montreal PQ H3T 1J4 Canada Univ Montreal Math & Stat Dept Montreal PQ H3T 1J4 Canada 

出 版 物：《KNOWLEDGE-BASED SYSTEMS》 (Knowl Based Syst)

年 卷 期：2025年第311卷

核心收录：

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Natural Sciences and Engineering Research Council of Canada, Canada (NSERC [RGPIN-2023-04370] TransMedTech Institute, Canada [Projet-0131-AI-DBS] Fonds de recherche du Quebec-Nature et technologies, Canada (FRQNT) [2022-2023-B1X-318568] Institut de valorisation des donnees (IVADO) [MSc-2022-0961371377] 

主　　题：Bayesian optimization Domain knowledge integration Prior-weighted acquisition function Region of interest Human-in-the-loop 

摘      要：In diverse fields of application, Bayesian Optimization (BO) has been proposed to find the optimum of black- box functions, surpassing human-driven searches. BO s appeal lies in its data efficiency, making it suitable for optimizing costly-to-evaluate functions without requiring extensive training data. While BO can perform well in closed-loop, domain experts frequently have hypotheses about which parameter combinations are more likely to yield optimal results. Hence, for BO to be truly relevant and adopted by practitioners, such prior knowledge needs to be efficiently and seamlessly integrated into the optimization framework. Some methods were recently developed to address this challenge, but they suffer from robustness issues when provided erroneous insight. Building on the idea of element-wise prior-weighted acquisition function, we propose to use a fixed-weight effective prior that distills expert user knowledge with minimal computational cost. Comprehensive investigation across diverse task conditions and prior quality levels revealed that our method, alpha- t BO, surpasses Vanilla BO when provided with insights of good quality while maintaining robustness against misleading information. Moreover, unlike other methods, alpha- t BO typically requires no hyperparameter tuning, largely simplifying its implementation in diverse tasks.