Actively Learning Gaussian Process Dynamics

Mona Buisson-Fenet, Friedrich Solowjow, Sebastian Trimpe
Proceedings of the 2nd Conference on Learning for Dynamics and Control, PMLR 120:5-15, 2020.

Abstract

Despite the availability of ever more data enabled through modern sensor and computer technology, it still remains an open problem to learn dynamical systems in a sample-efficient way.We propose active learning strategies that leverage information-theoretical properties arising naturally during Gaussian process regression, while respecting constraints on the sampling process imposed by the system dynamics. Sample points are selected in regions with high uncertainty, leading to exploratory behavior and data-efficient training of the model.All results are verified in an extensive numerical benchmark.

Cite this Paper


BibTeX
@InProceedings{pmlr-v120-buisson-fenet20a, title = {Actively Learning Gaussian Process Dynamics}, author = {Buisson-Fenet, Mona and Solowjow, Friedrich and Trimpe, Sebastian}, booktitle = {Proceedings of the 2nd Conference on Learning for Dynamics and Control}, pages = {5--15}, year = {2020}, editor = {Bayen, Alexandre M. and Jadbabaie, Ali and Pappas, George and Parrilo, Pablo A. and Recht, Benjamin and Tomlin, Claire and Zeilinger, Melanie}, volume = {120}, series = {Proceedings of Machine Learning Research}, month = {10--11 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v120/buisson-fenet20a/buisson-fenet20a.pdf}, url = {https://proceedings.mlr.press/v120/buisson-fenet20a.html}, abstract = {Despite the availability of ever more data enabled through modern sensor and computer technology, it still remains an open problem to learn dynamical systems in a sample-efficient way.We propose active learning strategies that leverage information-theoretical properties arising naturally during Gaussian process regression, while respecting constraints on the sampling process imposed by the system dynamics. Sample points are selected in regions with high uncertainty, leading to exploratory behavior and data-efficient training of the model.All results are verified in an extensive numerical benchmark.} }
Endnote
%0 Conference Paper %T Actively Learning Gaussian Process Dynamics %A Mona Buisson-Fenet %A Friedrich Solowjow %A Sebastian Trimpe %B Proceedings of the 2nd Conference on Learning for Dynamics and Control %C Proceedings of Machine Learning Research %D 2020 %E Alexandre M. Bayen %E Ali Jadbabaie %E George Pappas %E Pablo A. Parrilo %E Benjamin Recht %E Claire Tomlin %E Melanie Zeilinger %F pmlr-v120-buisson-fenet20a %I PMLR %P 5--15 %U https://proceedings.mlr.press/v120/buisson-fenet20a.html %V 120 %X Despite the availability of ever more data enabled through modern sensor and computer technology, it still remains an open problem to learn dynamical systems in a sample-efficient way.We propose active learning strategies that leverage information-theoretical properties arising naturally during Gaussian process regression, while respecting constraints on the sampling process imposed by the system dynamics. Sample points are selected in regions with high uncertainty, leading to exploratory behavior and data-efficient training of the model.All results are verified in an extensive numerical benchmark.
APA
Buisson-Fenet, M., Solowjow, F. & Trimpe, S.. (2020). Actively Learning Gaussian Process Dynamics. Proceedings of the 2nd Conference on Learning for Dynamics and Control, in Proceedings of Machine Learning Research 120:5-15 Available from https://proceedings.mlr.press/v120/buisson-fenet20a.html.

Related Material