Institute Homepage
Institute Homepage DE Sign In

Back

Autonomous Motion Intelligent Control Systems Conference Paper 2018

Probabilistic Recurrent State-Space Models

Thumb ticker sm doerr andreas 09 croped
Intelligent Control Systems
Andreas Doerr
Thumb ticker sm ss
Autonomous Motion
Stefan Schaal
  • Director
Thumb ticker sm 2018 ac r7b9314 cut
Intelligent Control Systems
Sebastian Trimpe
Teaser image

State-space models (SSMs) are a highly expressive model class for learning patterns in time series data and for system identification. Deterministic versions of SSMs (e.g., LSTMs) proved extremely successful in modeling complex time-series data. Fully probabilistic SSMs, however, unfortunately often prove hard to train, even for smaller problems. To overcome this limitation, we propose a scalable initialization and training algorithm based on doubly stochastic variational inference and Gaussian processes. In the variational approximation we propose in contrast to related approaches to fully capture the latent state temporal correlations to allow for robust training.

Author(s): Andreas Doerr and Christian Daniel and Martin Schiegg and Duy Nguyen-Tuong and Stefan Schaal and Marc Toussaint and Sebastian Trimpe
Book Title: Proceedings of the International Conference on Machine Learning (ICML)
Year: 2018
Month: July
Project(s):
Bibtex Type: Conference Paper (inproceedings)
Event Name: International Conference on Machine Learning (ICML)
Event Place: Stockholm, Sweden
State: Published
Electronic Archiving: grant_archive
Links:
Attachments:

BibTex 

@inproceedings{doerr2018probabilistic,
  title = {Probabilistic Recurrent State-Space Models},
  booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
  abstract = {State-space models (SSMs) are a highly expressive model class for learning patterns in time series data and for system identification. Deterministic versions of SSMs (e.g., LSTMs) proved extremely successful in modeling complex time-series data. Fully probabilistic SSMs, however, unfortunately often prove hard to train, even for smaller problems. To overcome this limitation, we propose a scalable initialization and training algorithm based on doubly stochastic variational inference and Gaussian processes. In the variational approximation we propose in contrast to related approaches to fully capture the latent state temporal correlations to allow for robust training. },
  month = jul,
  year = {2018},
  slug = {probabilistic-recurrent-state-space-models},
  author = {Doerr, Andreas and Daniel, Christian and Schiegg, Martin and Nguyen-Tuong, Duy and Schaal, Stefan and Toussaint, Marc and Trimpe, Sebastian},
  month_numeric = {7}
}