Renyi Entropy Search for Bayesian Optimization

Bayesian optimization (BO) offers a solution to intractable optimization problems. Exploration and exploitation (E&E) are determined in BO using acquisition functions, in particular by entropic search. Entropy search acquisition functions aim to reduce the maximum uncertainty in the problem solution space. This preliminary work presents an approach using Renyi's a entropy criterion for dynamic exploration. We combine the entropy search in certain and uncertain spaces on the predicted output Gaussian distribution. Our results support the interest of studying Renyi's a for E&E in entropic search.