Acquisition functions for simultaneous Bayesian optimisation of multiple problems
Michael Pearce · 2017
Using Gaussian Processes as statistical predictors of expensive objective functions for optimization has gathered much attention over the last two decades. Many acquisition functions that guide the search for data collection have been developed for various optimization cases, including multi-fidelity and multiple objectives. We look at the case where there are multiple tasks to be solved simultaneously. One example may be the algorithm selection problem where each use case of an algorithm is a unique optimization problem, and a user aims to find the optimal setting for each use case based on optimal settings of similar cases. Input uncertainty can be seen as a related problem where one must use the same settings for all use cases. We propose a variety of acquisition functions for Bayesian optimization in this general class of optimization problems.