The document describes a machine learning method for efficient design optimization in nano-optics using Gaussian process regression and Bayesian optimization. It discusses how Gaussian process regression can be used to build regression models from expensive black-box functions to enable model-based optimization and integration. Bayesian optimization is then used to iteratively query the black-box function at points of maximum expected improvement to find its minimum. The method can incorporate derivative observations to speed up optimization by providing additional training data for the Gaussian process. Differential evolution is also utilized to efficiently maximize the expected improvement at each iteration. The approach is demonstrated on benchmark optimization problems, showing it outperforms other algorithms like L-BFGS and particle swarm optimization.