[00912] Simulation-based Bayesian optimization over categorical covariates
Session Time & Room : 3C (Aug.23, 13:20-15:00) @A206
Type : Contributed Talk
Abstract : Optimizing black-box functions of categorical variables has important applications, including the design of biological sequences with specific properties. Bayesian optimization is widely used in this type of problem. It involves adjusting a probabilistic machine learning model of the objective and using an acquisition function to guide the optimization process. We propose a new algorithm to sequentially optimize the acquisition function inspired in simulated annealing. We address convergence issues and demonstrate its effectiveness on RNA-sequence optimization.