arXiv:2205.14090v2 Announce Type: replace 
Abstract: Bayesian optimization (BO) has well-documented merits for optimizing black-box functions with an expensive evaluation cost. Such functions emerge in applications as diverse as hyperparameter tuning, drug discovery, and robotics. BO hinges on a Bayesian surrogate model to sequentially select query points so as to balance exploration with exploitation of the search space. Most existing works rely on a single Gaussian process (GP) based surrogate model, where the kernel function form is typically preselected using domain knowledge. To bypass such a design process, this paper leverages an ensemble (E) of GPs to adaptively select the surrogate model fit on-the-fly, yielding a GP mixture posterior with enhanced expressiveness for the sought function. Acquisition of the next evaluation input using this EGP-based function posterior is then enabled by Thompson sampling (TS) that requires no additional design parameters. To endow function sampling with scalability, random feature-based kernel approximation is leveraged per GP model. The novel EGP-TS readily accommodates parallel operation. To further establish convergence of the proposed EGP-TS to the global optimum, analysis is conducted based on the notion of Bayesian regret for both sequential and parallel settings. Tests on synthetic functions and real-world applications showcase the merits of the proposed method.

قدمت دراسة جديدة نهجًا لنمذجة البدائل في التحسين البايزي يستخدم مجموعة من العمليات الغاوسية (GP) بدلاً من الاعتماد على GP واحد. تتيح هذه الطريقة اختيار النموذج البديل بشكل تكيفي، مما يعزز من تعبير الدالة التي يتم تحسينها. يتم تسهيل الحصول على نقطة التقييم التالية من خلال أخذ عينات تومسون، التي لا تتطلب معلمات تصميم إضافية.

Un nuevo estudio ha presentado un enfoque de modelado de sustitución para la optimización bayesiana que utiliza un conjunto de procesos gaussianos (GP) en lugar de depender de un solo GP. Este método permite la selección adaptativa del modelo de sustitución, mejorando la expresividad de la función que se está optimizando. La adquisición del siguiente punto de evaluación se facilita a través del muestreo de Thompson, que no requiere parámetros de diseño adicionales.

Une nouvelle étude a introduit une approche de modélisation de substitution pour l'optimisation bayésienne qui utilise un ensemble de processus gaussiens (GP) au lieu de s'appuyer sur un seul GP. Cette méthode permet une sélection adaptative du modèle de substitution, améliorant ainsi l'expressivité de la fonction optimisée. L'acquisition du prochain point d'évaluation est facilitée par l'échantillonnage de Thompson, qui ne nécessite pas de paramètres de conception supplémentaires.

A new study has introduced a surrogate modeling approach for Bayesian optimization that utilizes an ensemble of Gaussian processes (GPs) instead of relying on a single GP. This method allows for adaptive selection of the surrogate model, enhancing the expressiveness of the function being optimized. The acquisition of the next evaluation input is facilitated through Thompson sampling, which does not require additional design parameters.