Spatio-temporal neural networks for space-time data modeling and relation discovery

Abstract : We introduce a dynamical spatio-temporal model formalized as a recurrent neural network for modeling time series of spatial processes, i.e. series of observations sharing temporal and spatial dependencies. The model learns these dependencies through a structured latent dynamical component, while a decoder predicts the observations from the latent representations. We consider several variants of this model, corresponding to different prior hypothesis about the spatial relations between the series. The model is used for the tasks of forecasting and data imputation. It is evaluated and compared to state-of-the-art baselines, on a variety of forecasting and imputation problems representative of different application areas: epidemiology, geo-spatial statistics and car-traffic prediction. The experiments also show that this approach is able to learn relevant spatial relations without prior information.
Document type :
Journal articles
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-02336358
Contributor : Edouard Delasalles <>
Submitted on : Monday, October 28, 2019 - 5:40:57 PM
Last modification on : Wednesday, October 30, 2019 - 1:34:49 AM

Identifiers

Citation

Edouard Delasalles, Ali Ziat, Ludovic Denoyer, Patrick Gallinari. Spatio-temporal neural networks for space-time data modeling and relation discovery. Knowledge and Information Systems (KAIS), Springer, 2019, 61 (3), pp.1241-1267. ⟨10.1007/s10115-018-1291-x⟩. ⟨hal-02336358⟩

Share

Metrics

Record views

26