%0 Generic %1 karlbauer2020hidden %A Karlbauer, Matthias %A Menge, Tobias %A Otte, Sebastian %A Lensch, Hendrik P. A. %A Scholten, Thomas %A Wulfmeyer, Volker %A Butz, Martin V. %D 2020 %K climate climatechange climatemodeling spacialtemporal todo:read %T Hidden Latent State Inference in a Spatio-Temporal Generative Model %U http://arxiv.org/abs/2009.09823 %X Knowledge of the hidden factors that determine particular system dynamics is crucial for both explaining them and pursuing goal-directed, interventional actions. The inference of these factors without supervision given time series data remains an open challenge. Here, we focus on spatio-temporal processes, including wave propagations and weather dynamics, and assume that universal causes (e.g. physics) apply throughout space and time. We apply a novel DIstributed, Spatio-Temporal graph Artificial Neural network Architecture, DISTANA, which learns a generative model in such domains. DISTANA requires fewer parameters, and yields more accurate predictions than temporal convolutional neural networks and other related approaches on a 2D circular wave prediction task. We show that DISTANA, when combined with a retrospective latent state inference principle called active tuning, can reliably derive hidden local causal factors. In a current weather prediction benchmark, DISTANA infers our planet's land-sea mask solely by observing temperature dynamics and uses the self inferred information to improve its own prediction of temperature. We are convinced that the retrospective inference of latent states in generative RNN architectures will play an essential role in future research on causal inference and explainable systems.

@misc{karlbauer2020hidden, abstract = {Knowledge of the hidden factors that determine particular system dynamics is crucial for both explaining them and pursuing goal-directed, interventional actions. The inference of these factors without supervision given time series data remains an open challenge. Here, we focus on spatio-temporal processes, including wave propagations and weather dynamics, and assume that universal causes (e.g. physics) apply throughout space and time. We apply a novel DIstributed, Spatio-Temporal graph Artificial Neural network Architecture, DISTANA, which learns a generative model in such domains. DISTANA requires fewer parameters, and yields more accurate predictions than temporal convolutional neural networks and other related approaches on a 2D circular wave prediction task. We show that DISTANA, when combined with a retrospective latent state inference principle called active tuning, can reliably derive hidden local causal factors. In a current weather prediction benchmark, DISTANA infers our planet's land-sea mask solely by observing temperature dynamics and uses the self inferred information to improve its own prediction of temperature. We are convinced that the retrospective inference of latent states in generative RNN architectures will play an essential role in future research on causal inference and explainable systems.}, added-at = {2021-02-16T11:51:17.000+0100}, author = {Karlbauer, Matthias and Menge, Tobias and Otte, Sebastian and Lensch, Hendrik P. A. and Scholten, Thomas and Wulfmeyer, Volker and Butz, Martin V.}, biburl = {https://www.bibsonomy.org/bibtex/25fa27c868257069941da095b442321ba/annakrause}, description = {2009.09823v1.pdf}, interhash = {5711e126980308d525abfb0445867e50}, intrahash = {5fa27c868257069941da095b442321ba}, keywords = {climate climatechange climatemodeling spacialtemporal todo:read}, note = {cite arxiv:2009.09823Comment: As submitted to the 35th conference of the Association for the Advancement of Artificial Intelligence (AAAI-21)}, timestamp = {2021-02-16T11:51:17.000+0100}, title = {Hidden Latent State Inference in a Spatio-Temporal Generative Model}, url = {http://arxiv.org/abs/2009.09823}, year = 2020 }