Gaussian Process Regression with GPyTorch

So far, we have grasped the main components of a Bayesian optimization procedure: a surrogate model that provides posterior estimates on the mean and uncertainty of the underlying objective function and an acquisition function that guides the search for the next sampling location based on its expected gain in the marginal utility. Efficiently calculating the posterior distributions becomes essential in the case of parallel Bayesian optimization and Monte Carlo acquisition functions. This branch evaluates multiple points simultaneously discussed in a later chapter.