Intelligent Control Systems Conference Paper 2020

Actively Learning Gaussian Process Dynamics

L4dc legend

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.

Author(s): Mona Buisson-Fenet and Friedrich Solowjow and Sebastian Trimpe
Book Title: Proceedings of the 2nd Conference on Learning for Dynamics and Control
Volume: 120
Pages: 5--15
Year: 2020
Month: June
Series: Proceedings of Machine Learning Research (PMLR)
Editors: Bayen, Alexandre M. and Jadbabaie, Ali and Pappas, George and Parrilo, Pablo A. and Recht, Benjamin and Tomlin, Claire and Zeilinger, Melanie
Publisher: PMLR
Bibtex Type: Conference Paper (conference)
Event Name: 2nd Annual Conference on Learning for Dynamics and Control (L4DC)
Event Place: The Cloud
State: Published
URL: http://proceedings.mlr.press/v120/buisson-fenet20a.html
Electronic Archiving: grant_archive
Links:
Attachments:

BibTex

@conference{Actively_Learning_Gaussian_Process_Dynamics,
  title = {Actively Learning Gaussian Process Dynamics},
  booktitle = {Proceedings of the 2nd Conference on Learning for Dynamics and Control},
  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.
  },
  volume = {120},
  pages = {5--15},
  series = {Proceedings of Machine Learning Research (PMLR)},
  editors = {Bayen, Alexandre M. and Jadbabaie, Ali and Pappas, George and Parrilo, Pablo A. and Recht, Benjamin and Tomlin, Claire and Zeilinger, Melanie},
  publisher = {PMLR},
  month = jun,
  year = {2020},
  slug = {actively-learning-gaussian-process-dynamics},
  author = {Buisson-Fenet, Mona and Solowjow, Friedrich and Trimpe, Sebastian},
  url = {http://proceedings.mlr.press/v120/buisson-fenet20a.html},
  month_numeric = {6}
}