Meta-Learning Priors for Safe Bayesian Optimization

Jonas Rothfuss, Christopher Koenig, Alisa Rupenyan, Andreas Krause
Proceedings of The 6th Conference on Robot Learning, PMLR 205:237-265, 2023.

Abstract

In robotics, optimizing controller parameters under safety constraints is an important challenge. Safe Bayesian optimization (BO) quantifies uncertainty in the objective and constraints to safely guide exploration in such settings. Hand-designing a suitable probabilistic model can be challenging however. In the presence of unknown safety constraints, it is crucial to choose reliable model hyper-parameters to avoid safety violations. Here, we propose a data-driven approach to this problem by em meta-learning priors for safe BO from offline data. We build on a meta-learning algorithm, F-PACOH, capable of providing reliable uncertainty quantification in settings of data scarcity. As core contribution, we develop a novel framework for choosing safety-compliant priors in a data-riven manner via empirical uncertainty metrics and a frontier search algorithm. On benchmark functions and a high-precision motion system, we demonstrate that our meta-learnt priors accelerate convergence of safe BO approaches while maintaining safety.

Cite this Paper

BibTeX
@InProceedings{pmlr-v205-rothfuss23a, title = {Meta-Learning Priors for Safe Bayesian Optimization}, author = {Rothfuss, Jonas and Koenig, Christopher and Rupenyan, Alisa and Krause, Andreas}, booktitle = {Proceedings of The 6th Conference on Robot Learning}, pages = {237--265}, year = {2023}, editor = {Liu, Karen and Kulic, Dana and Ichnowski, Jeff}, volume = {205}, series = {Proceedings of Machine Learning Research}, month = {14--18 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v205/rothfuss23a/rothfuss23a.pdf}, url = {https://proceedings.mlr.press/v205/rothfuss23a.html}, abstract = {In robotics, optimizing controller parameters under safety constraints is an important challenge. Safe Bayesian optimization (BO) quantifies uncertainty in the objective and constraints to safely guide exploration in such settings. Hand-designing a suitable probabilistic model can be challenging however. In the presence of unknown safety constraints, it is crucial to choose reliable model hyper-parameters to avoid safety violations. Here, we propose a data-driven approach to this problem by em meta-learning priors for safe BO from offline data. We build on a meta-learning algorithm, F-PACOH, capable of providing reliable uncertainty quantification in settings of data scarcity. As core contribution, we develop a novel framework for choosing safety-compliant priors in a data-riven manner via empirical uncertainty metrics and a frontier search algorithm. On benchmark functions and a high-precision motion system, we demonstrate that our meta-learnt priors accelerate convergence of safe BO approaches while maintaining safety. } }
Endnote
%0 Conference Paper %T Meta-Learning Priors for Safe Bayesian Optimization %A Jonas Rothfuss %A Christopher Koenig %A Alisa Rupenyan %A Andreas Krause %B Proceedings of The 6th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2023 %E Karen Liu %E Dana Kulic %E Jeff Ichnowski %F pmlr-v205-rothfuss23a %I PMLR %P 237--265 %U https://proceedings.mlr.press/v205/rothfuss23a.html %V 205 %X In robotics, optimizing controller parameters under safety constraints is an important challenge. Safe Bayesian optimization (BO) quantifies uncertainty in the objective and constraints to safely guide exploration in such settings. Hand-designing a suitable probabilistic model can be challenging however. In the presence of unknown safety constraints, it is crucial to choose reliable model hyper-parameters to avoid safety violations. Here, we propose a data-driven approach to this problem by em meta-learning priors for safe BO from offline data. We build on a meta-learning algorithm, F-PACOH, capable of providing reliable uncertainty quantification in settings of data scarcity. As core contribution, we develop a novel framework for choosing safety-compliant priors in a data-riven manner via empirical uncertainty metrics and a frontier search algorithm. On benchmark functions and a high-precision motion system, we demonstrate that our meta-learnt priors accelerate convergence of safe BO approaches while maintaining safety.
APA
Rothfuss, J., Koenig, C., Rupenyan, A. & Krause, A.. (2023). Meta-Learning Priors for Safe Bayesian Optimization. Proceedings of The 6th Conference on Robot Learning, in Proceedings of Machine Learning Research 205:237-265 Available from https://proceedings.mlr.press/v205/rothfuss23a.html.

Related Material