Efficient Rollout Strategies for Bayesian Optimization

Eric Lee, David Eriksson, David Bindel, Bolong Cheng, Mike Mccourt
Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI), PMLR 124:260-269, 2020.

Abstract

Bayesian optimization (BO) is a class of sample-efficient global optimization methods, where a probabilistic model conditioned on previous observations is used to determine future evaluations via the optimization of an acquisition function. Most acquisition functions are myopic, meaning that they only consider the impact of the next function evaluation. Non-myopic acquisition functions consider the impact of the next h function evaluations and are typically computed through rollout, in which h steps of BO are simulated. These rollout acquisition functions are defined as h-dimensional integrals, and are expensive to compute and optimize. We show that a combination of quasi-Monte Carlo, common random numbers, and control variates significantly reduce the computational burden of rollout. We then formulate a policy-search based approach that removes the need to optimize the rollout acquisition function. Finally, we discuss the qualitative behavior of rollout policies in the setting of multi-modal objectives and model error.

Cite this Paper


BibTeX
@InProceedings{pmlr-v124-lee20a, title = {Efficient Rollout Strategies for Bayesian Optimization}, author = {Lee, Eric and Eriksson, David and Bindel, David and Cheng, Bolong and Mccourt, Mike}, booktitle = {Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI)}, pages = {260--269}, year = {2020}, editor = {Jonas Peters and David Sontag}, volume = {124}, series = {Proceedings of Machine Learning Research}, month = {03--06 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v124/lee20a/lee20a.pdf}, url = { http://proceedings.mlr.press/v124/lee20a.html }, abstract = {Bayesian optimization (BO) is a class of sample-efficient global optimization methods, where a probabilistic model conditioned on previous observations is used to determine future evaluations via the optimization of an acquisition function. Most acquisition functions are myopic, meaning that they only consider the impact of the next function evaluation. Non-myopic acquisition functions consider the impact of the next h function evaluations and are typically computed through rollout, in which h steps of BO are simulated. These rollout acquisition functions are defined as h-dimensional integrals, and are expensive to compute and optimize. We show that a combination of quasi-Monte Carlo, common random numbers, and control variates significantly reduce the computational burden of rollout. We then formulate a policy-search based approach that removes the need to optimize the rollout acquisition function. Finally, we discuss the qualitative behavior of rollout policies in the setting of multi-modal objectives and model error.} }
Endnote
%0 Conference Paper %T Efficient Rollout Strategies for Bayesian Optimization %A Eric Lee %A David Eriksson %A David Bindel %A Bolong Cheng %A Mike Mccourt %B Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI) %C Proceedings of Machine Learning Research %D 2020 %E Jonas Peters %E David Sontag %F pmlr-v124-lee20a %I PMLR %P 260--269 %U http://proceedings.mlr.press/v124/lee20a.html %V 124 %X Bayesian optimization (BO) is a class of sample-efficient global optimization methods, where a probabilistic model conditioned on previous observations is used to determine future evaluations via the optimization of an acquisition function. Most acquisition functions are myopic, meaning that they only consider the impact of the next function evaluation. Non-myopic acquisition functions consider the impact of the next h function evaluations and are typically computed through rollout, in which h steps of BO are simulated. These rollout acquisition functions are defined as h-dimensional integrals, and are expensive to compute and optimize. We show that a combination of quasi-Monte Carlo, common random numbers, and control variates significantly reduce the computational burden of rollout. We then formulate a policy-search based approach that removes the need to optimize the rollout acquisition function. Finally, we discuss the qualitative behavior of rollout policies in the setting of multi-modal objectives and model error.
APA
Lee, E., Eriksson, D., Bindel, D., Cheng, B. & Mccourt, M.. (2020). Efficient Rollout Strategies for Bayesian Optimization. Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence (UAI), in Proceedings of Machine Learning Research 124:260-269 Available from http://proceedings.mlr.press/v124/lee20a.html .

Related Material