Spatio-Temporal Neural Networks for Space-Time Series Forecasting and Relations Discovery

Abstract

We introduce a dynamical spatio-temporal model formalized as a recurrent neural network for forecasting 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 evaluated and compared to state-of-the-art baselines, on a variety of forecasting problems representative of different application areas: epidemiology, geo-spatial statistics and car-traffic prediction. Besides these evaluations, we also describe experiments showing the ability of this approach to extract relevant spatial relations.

arXiv commentaccepted by: ICDM 2018 - IEEE International Conference on Data Mining series (ICDM)

References in corpus (7)

• Auto-Encoding Variational Bayes
• Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting
• Unsupervised Learning of Video Representations using LSTMs
• A Recurrent Latent Variable Model for Sequential Data
• Gated Graph Sequence Neural Networks
• Learning Stochastic Recurrent Networks
• Deep Kalman Filters

Cited by in corpus (17)

• Deep learning for time series classification: a review
• Deep Learning on Traffic Prediction: Methods, Analysis and Future Directions
• Transfer learning for time series classification
• Learning Dynamical Systems from Partial Observations
• Forecasting the outcome of spintronic experiments with Neural Ordinary Differential Equations
• Forecaster: A Graph Transformer for Forecasting Spatial and Time-Dependent Data
• Demystifying Deep Learning in Predictive Spatio-Temporal Analytics: An Information-Theoretic Framework
• Deep learning for time series classification
• FedST: Secure Federated Shapelet Transformation for Time Series Classification
• Live Multi-Streaming and Donation Recommendations via Coupled Donation-Response Tensor Factorization
• Adaptive Transfer Learning of Multi-View Time Series Classification
• Spatiotemporal Adaptive Neural Network for Long-term Forecasting of Financial Time Series
• Deep Geospatial Interpolation Networks
• GSA-Forecaster: Forecasting Graph-Based Time-Dependent Data with Graph Sequence Attention
• A Context Integrated Relational Spatio-Temporal Model for Demand and Supply Forecasting
• Piece-wise Matching Layer in Representation Learning for ECG Classification
• FeatureNorm: L2 Feature Normalization for Dynamic Graph Embedding