Minimum Regret Search for Single- and Multi-Task Optimization

Jan Hendrik Metzen
Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:192-200, 2016.

Abstract

We propose minimum regret search (MRS), a novel acquisition function for Bayesian optimization. MRS bears similarities with information-theoretic approaches such as entropy search (ES). However, while ES aims in each query at maximizing the information gain with respect to the global maximum, MRS aims at minimizing the expected simple regret of its ultimate recommendation for the optimum. While empirically ES and MRS perform similar in most of the cases, MRS produces fewer outliers with high simple regret than ES. We provide empirical results both for a synthetic single-task optimization problem as well as for a simulated multi-task robotic control problem.

Cite this Paper

BibTeX
@InProceedings{pmlr-v48-metzen16, title = {Minimum Regret Search for Single- and Multi-Task Optimization}, author = {Metzen, Jan Hendrik}, booktitle = {Proceedings of The 33rd International Conference on Machine Learning}, pages = {192--200}, year = {2016}, editor = {Balcan, Maria Florina and Weinberger, Kilian Q.}, volume = {48}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {20--22 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v48/metzen16.pdf}, url = {https://proceedings.mlr.press/v48/metzen16.html}, abstract = {We propose minimum regret search (MRS), a novel acquisition function for Bayesian optimization. MRS bears similarities with information-theoretic approaches such as entropy search (ES). However, while ES aims in each query at maximizing the information gain with respect to the global maximum, MRS aims at minimizing the expected simple regret of its ultimate recommendation for the optimum. While empirically ES and MRS perform similar in most of the cases, MRS produces fewer outliers with high simple regret than ES. We provide empirical results both for a synthetic single-task optimization problem as well as for a simulated multi-task robotic control problem.} }
Endnote
%0 Conference Paper %T Minimum Regret Search for Single- and Multi-Task Optimization %A Jan Hendrik Metzen %B Proceedings of The 33rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Maria Florina Balcan %E Kilian Q. Weinberger %F pmlr-v48-metzen16 %I PMLR %P 192--200 %U https://proceedings.mlr.press/v48/metzen16.html %V 48 %X We propose minimum regret search (MRS), a novel acquisition function for Bayesian optimization. MRS bears similarities with information-theoretic approaches such as entropy search (ES). However, while ES aims in each query at maximizing the information gain with respect to the global maximum, MRS aims at minimizing the expected simple regret of its ultimate recommendation for the optimum. While empirically ES and MRS perform similar in most of the cases, MRS produces fewer outliers with high simple regret than ES. We provide empirical results both for a synthetic single-task optimization problem as well as for a simulated multi-task robotic control problem.
RIS
TY - CPAPER TI - Minimum Regret Search for Single- and Multi-Task Optimization AU - Jan Hendrik Metzen BT - Proceedings of The 33rd International Conference on Machine Learning DA - 2016/06/11 ED - Maria Florina Balcan ED - Kilian Q. Weinberger ID - pmlr-v48-metzen16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 48 SP - 192 EP - 200 L1 - http://proceedings.mlr.press/v48/metzen16.pdf UR - https://proceedings.mlr.press/v48/metzen16.html AB - We propose minimum regret search (MRS), a novel acquisition function for Bayesian optimization. MRS bears similarities with information-theoretic approaches such as entropy search (ES). However, while ES aims in each query at maximizing the information gain with respect to the global maximum, MRS aims at minimizing the expected simple regret of its ultimate recommendation for the optimum. While empirically ES and MRS perform similar in most of the cases, MRS produces fewer outliers with high simple regret than ES. We provide empirical results both for a synthetic single-task optimization problem as well as for a simulated multi-task robotic control problem. ER -
APA
Metzen, J.H.. (2016). Minimum Regret Search for Single- and Multi-Task Optimization. Proceedings of The 33rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 48:192-200 Available from https://proceedings.mlr.press/v48/metzen16.html.

Related Material