Collapsed Amortized Variational Inference for Switching Nonlinear Dynamical Systems

Zhe Dong, Bryan Seybold, Kevin Murphy, Hung Bui
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:2638-2647, 2020.

Abstract

We propose an efficient inference method for switching nonlinear dynamical systems. The key idea is to learn an inference network which can be used as a proposal distribution for the continuous latent variables, while performing exact marginalization of the discrete latent variables. This allows us to use the reparameterization trick, and apply end-to-end training with stochastic gradient descent. We show that the proposed method can successfully segment time series data, including videos and 3D human pose, into meaningful “regimes” by using the piece-wise nonlinear dynamics.

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-dong20e, title = {Collapsed Amortized Variational Inference for Switching Nonlinear Dynamical Systems}, author = {Dong, Zhe and Seybold, Bryan and Murphy, Kevin and Bui, Hung}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {2638--2647}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/dong20e/dong20e.pdf}, url = {https://proceedings.mlr.press/v119/dong20e.html}, abstract = {We propose an efficient inference method for switching nonlinear dynamical systems. The key idea is to learn an inference network which can be used as a proposal distribution for the continuous latent variables, while performing exact marginalization of the discrete latent variables. This allows us to use the reparameterization trick, and apply end-to-end training with stochastic gradient descent. We show that the proposed method can successfully segment time series data, including videos and 3D human pose, into meaningful “regimes” by using the piece-wise nonlinear dynamics.} }
Endnote
%0 Conference Paper %T Collapsed Amortized Variational Inference for Switching Nonlinear Dynamical Systems %A Zhe Dong %A Bryan Seybold %A Kevin Murphy %A Hung Bui %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-dong20e %I PMLR %P 2638--2647 %U https://proceedings.mlr.press/v119/dong20e.html %V 119 %X We propose an efficient inference method for switching nonlinear dynamical systems. The key idea is to learn an inference network which can be used as a proposal distribution for the continuous latent variables, while performing exact marginalization of the discrete latent variables. This allows us to use the reparameterization trick, and apply end-to-end training with stochastic gradient descent. We show that the proposed method can successfully segment time series data, including videos and 3D human pose, into meaningful “regimes” by using the piece-wise nonlinear dynamics.
APA
Dong, Z., Seybold, B., Murphy, K. & Bui, H.. (2020). Collapsed Amortized Variational Inference for Switching Nonlinear Dynamical Systems. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:2638-2647 Available from https://proceedings.mlr.press/v119/dong20e.html.

Related Material