Acquisition
Acquisition describes the process of picking a new set of points to evaluate, given the previously fitted surrogate model.
For this process, three types of methods are considered.
Utility Function
Given a surrogate model, the utility function returns a single point to evaluate next.
Random
Optimize a random sample
Simple Regret
Optimize input samples
Uncertainty Utility
Posterior Variance, optimize deviation from mean
Quantile Utility
Upper Confidence Bound
Quantitative Improvement
Expected Improvement
Qualitative Improvement
Probability of Improvement
Acquisition Strategy
Given a utility function, the acquisition strategy determines how the utility function is called to return a single point to evaluate next.
Sequential Acquisition
Call the utility function once to determine the next point (1LA)
Sequential Lookahead Acquisition
Use the utility function, apply its result and use the utility function again (2LA)
Sequential Joint Acquisition
Joint optimization (2JointLA)
Batching Strategy
Given an acquisition strategy, the batching strategy returns points, each of which are picked by some acquisition strategy.