Definition
A classical search method that picks the next experiment based on a probabilistic model of past results.
A black-box optimization technique that fits a surrogate (often a Gaussian process) over an objective and uses an acquisition function to choose new query points; a traditional baseline for neural architecture search.