Preferential Bayesian Optimization

Preferential Bayesian Optimization (PBO) is a machine learning framework designed to efficiently optimize functions using only pairwise comparisons of candidate solutions, rather than direct function evaluations. Current research focuses on improving PBO's robustness and efficiency by addressing issues like multi-objective optimization, heteroscedastic (unequal) noise in preference data, and developing theoretically sound algorithms with provable convergence guarantees. These advancements are leading to more reliable and practical applications in diverse fields, such as robotics, autonomous driving, and materials science, where direct function evaluation may be expensive, time-consuming, or even impossible.