Switching Linear Dynamics for Variational Bayes Filtering

Philip Becker-Ehmck, Jan Peters, Patrick Van Der Smagt
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:553-562, 2019.

Abstract

System identification of complex and nonlinear systems is a central problem for model predictive control and model-based reinforcement learning. Despite their complexity, such systems can often be approximated well by a set of linear dynamical systems if broken into appropriate subsequences. This mechanism not only helps us find good approximations of dynamics, but also gives us deeper insight into the underlying system. Leveraging Bayesian inference, Variational Autoencoders and Concrete relaxations, we show how to learn a richer and more meaningful state space, e.g. encoding joint constraints and collisions with walls in a maze, from partial and high-dimensional observations. This representation translates into a gain of accuracy of learned dynamics showcased on various simulated tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-becker-ehmck19a, title = {Switching Linear Dynamics for Variational {B}ayes Filtering}, author = {Becker-Ehmck, Philip and Peters, Jan and Van Der Smagt, Patrick}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {553--562}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/becker-ehmck19a/becker-ehmck19a.pdf}, url = {https://proceedings.mlr.press/v97/becker-ehmck19a.html}, abstract = {System identification of complex and nonlinear systems is a central problem for model predictive control and model-based reinforcement learning. Despite their complexity, such systems can often be approximated well by a set of linear dynamical systems if broken into appropriate subsequences. This mechanism not only helps us find good approximations of dynamics, but also gives us deeper insight into the underlying system. Leveraging Bayesian inference, Variational Autoencoders and Concrete relaxations, we show how to learn a richer and more meaningful state space, e.g. encoding joint constraints and collisions with walls in a maze, from partial and high-dimensional observations. This representation translates into a gain of accuracy of learned dynamics showcased on various simulated tasks.} }
Endnote
%0 Conference Paper %T Switching Linear Dynamics for Variational Bayes Filtering %A Philip Becker-Ehmck %A Jan Peters %A Patrick Van Der Smagt %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-becker-ehmck19a %I PMLR %P 553--562 %U https://proceedings.mlr.press/v97/becker-ehmck19a.html %V 97 %X System identification of complex and nonlinear systems is a central problem for model predictive control and model-based reinforcement learning. Despite their complexity, such systems can often be approximated well by a set of linear dynamical systems if broken into appropriate subsequences. This mechanism not only helps us find good approximations of dynamics, but also gives us deeper insight into the underlying system. Leveraging Bayesian inference, Variational Autoencoders and Concrete relaxations, we show how to learn a richer and more meaningful state space, e.g. encoding joint constraints and collisions with walls in a maze, from partial and high-dimensional observations. This representation translates into a gain of accuracy of learned dynamics showcased on various simulated tasks.
APA
Becker-Ehmck, P., Peters, J. & Van Der Smagt, P.. (2019). Switching Linear Dynamics for Variational Bayes Filtering. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:553-562 Available from https://proceedings.mlr.press/v97/becker-ehmck19a.html.

Related Material