Local Bayesian Optimization

Local Bayesian optimization (LBO) tackles high-dimensional black-box function optimization by focusing on local search, thereby mitigating the "curse of dimensionality" that plagues global methods. Current research emphasizes improving local search strategies, such as minimizing uncertainty bounds or maximizing the probability of descent, often within the framework of Gaussian processes or random forests. These advancements aim to enhance the efficiency and convergence rates of Bayesian optimization, particularly in applications like hyperparameter tuning where the objective function landscape may exhibit favorable local structure. The resulting improvements in sample efficiency have significant implications for computationally expensive optimization problems across various scientific and engineering domains.