Disentangled Generative Models for Robust Prediction of System Dynamics

Stathi Fotiadis, Mario Lino Valencia, Shunlong Hu, Stef Garasto, Chris D Cantwell, Anil Anthony Bharath
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:10222-10248, 2023.

Abstract

The use of deep neural networks for modelling system dynamics is increasingly popular, but long-term prediction accuracy and out-of-distribution generalization still present challenges. In this study, we address these challenges by considering the parameters of dynamical systems as factors of variation of the data and leverage their ground-truth values to disentangle the representations learned by generative models. Our experimental results in phase-space and observation-space dynamics, demonstrate the effectiveness of latent-space supervision in producing disentangled representations, leading to improved long-term prediction accuracy and out-of-distribution robustness.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-fotiadis23a, title = {Disentangled Generative Models for Robust Prediction of System Dynamics}, author = {Fotiadis, Stathi and Lino Valencia, Mario and Hu, Shunlong and Garasto, Stef and Cantwell, Chris D and Bharath, Anil Anthony}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {10222--10248}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/fotiadis23a/fotiadis23a.pdf}, url = {https://proceedings.mlr.press/v202/fotiadis23a.html}, abstract = {The use of deep neural networks for modelling system dynamics is increasingly popular, but long-term prediction accuracy and out-of-distribution generalization still present challenges. In this study, we address these challenges by considering the parameters of dynamical systems as factors of variation of the data and leverage their ground-truth values to disentangle the representations learned by generative models. Our experimental results in phase-space and observation-space dynamics, demonstrate the effectiveness of latent-space supervision in producing disentangled representations, leading to improved long-term prediction accuracy and out-of-distribution robustness.} }
Endnote
%0 Conference Paper %T Disentangled Generative Models for Robust Prediction of System Dynamics %A Stathi Fotiadis %A Mario Lino Valencia %A Shunlong Hu %A Stef Garasto %A Chris D Cantwell %A Anil Anthony Bharath %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-fotiadis23a %I PMLR %P 10222--10248 %U https://proceedings.mlr.press/v202/fotiadis23a.html %V 202 %X The use of deep neural networks for modelling system dynamics is increasingly popular, but long-term prediction accuracy and out-of-distribution generalization still present challenges. In this study, we address these challenges by considering the parameters of dynamical systems as factors of variation of the data and leverage their ground-truth values to disentangle the representations learned by generative models. Our experimental results in phase-space and observation-space dynamics, demonstrate the effectiveness of latent-space supervision in producing disentangled representations, leading to improved long-term prediction accuracy and out-of-distribution robustness.
APA
Fotiadis, S., Lino Valencia, M., Hu, S., Garasto, S., Cantwell, C.D. & Bharath, A.A.. (2023). Disentangled Generative Models for Robust Prediction of System Dynamics. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:10222-10248 Available from https://proceedings.mlr.press/v202/fotiadis23a.html.

Related Material