Keywords: Bayesian optimization, meta-learning, learning to optimize
Abstract: In Bayesian optimisation, we often seek to minimise the black-box objective functions that arise in real-world physical systems. % ranging from engines to particle accelerators. A primary contributor to the cost of evaluating such black-box objective functions is often the effort required to prepare the system for measurement. We consider a common scenario where preparation costs grow as the distance between successive evaluations increases. %henceforth referred to as movement costs. In this setting, smooth optimisation trajectories are preferred and the jumpy paths produced by the standard myopic (i.e.\ one-step-optimal) Bayesian optimisation methods are sub-optimal. %However, existing non-myopic approaches do not support the long time-horizons required for path-wise smooth global optimisation. Our algorithm, MONGOOSE, uses a meta-learnt parametric policy to generate smooth optimisation trajectories, achieving performance gains over existing methods when optimising functions with large movement costs.
Submission Number: 44
Loading