Exploring Efficient and Simple Initialization Strategies for Bayesian Optimization with SETUP-BO

Seyed Ali YaghoubNejad, Mohammad Taghi Manzuri

Keywords: Bayesian optimization, initialization strategies, random initialization, grid initialization, optimization

TL;DR: This paper suggests using random and grid initialization strategies on SETUP-BO (i.e a variant of Bayesian optimization). The results show improvements in SETUP-BO with random initialization.

Abstract: This paper studies the effectiveness of random and grid initialization strategies in SETUP-BO, a self-tuning Bayesian optimization algorithm. Our experiments on benchmark functions compare the performance of these initialization strategies to deterministic initialization. The results show that random initialization outperforms other methods, indicating that it can enhance the performance of BO.