Batch Bayesian Optimization via Maximizing Variance Change

This paper proposes a new Bayesian optimization (BO) framework to avoid the unclear termination conditions in existing BO methods. Based on this framework, this paper uses Gaussian process as the surrogate model and constructs an acquisition function by minimizing the prediction error on the potential sample set, thus designing the batch BO algorithm via maximizing variance change (BOM algorithm). Simulation and experimental results show the effectiveness of the proposed BO framework and BOM algorithm.