Efficient spatio-temporal Gaussian regression via Kalman filtering

DOI10.1016/J.AUTOMATICA.2020.109032zbMATH Open1447.93357arXiv1705.01485OpenAlexW2610838890MaRDI QIDQ2188275FDOQ2188275

Authors: Marco Todescato, Andrea Carron, Ruggero Carli, Gianluigi Pillonetto, Luca Schenato 

Publication date: 10 June 2020

Published in: Automatica (Search for Journal in Brave)

Abstract: In this work we study the non-parametric reconstruction of spatio-temporal dynamical Gaussian processes (GPs) via GP regression from sparse and noisy data. GPs have been mainly applied to spatial regression where they represent one of the most powerful estimation approaches also thanks to their universal representing properties. Their extension to dynamical processes has been instead elusive so far since classical implementations lead to unscalable algorithms. We then propose a novel procedure to address this problem by coupling GP regression and Kalman filtering. In particular, assuming space/time separability of the covariance (kernel) of the process and rational time spectrum, we build a finite-dimensional discrete-time state-space process representation amenable of Kalman filtering. With sampling over a finite set of fixed spatial locations, our major finding is that the Kalman filter state at instant $t_k$ represents a sufficient statistic to compute the minimum variance estimate of the process at any $tgeq{t}_k$ over the entire spatial domain. This result can be interpreted as a novel Kalman representer theorem for dynamical GPs. We then extend the study to situations where the set of spatial input locations can vary over time. The proposed algorithms are finally tested on both synthetic and real field data, also providing comparisons with standard GP and truncated GP regression techniques.

Full work available at URL: https://arxiv.org/abs/1705.01485

Recommendations

• Recursive estimation for sparse Gaussian process regression
• An efficient implementation for spatial-temporal Gaussian process regression and its applications
• A dimension-reduced approach to space-time Kalman filtering
• Multilevel ensemble Kalman filtering for spatio-temporal processes
• A Kalman filter method for estimation and prediction of space-time data with an autoregressive structure

zbMATH Keywords

Kalman filtermachine learningrepresenter theoremseparable kernel functionsspatio-temporal Gaussian processes

Mathematics Subject Classification ID

Nonparametric regression and quantile regression (62G08) Gaussian processes (60G15) Filtering in stochastic control theory (93E11)

Cites Work

• Regularization networks and support vector machines
• Stochastic processes and filtering theory
• Gaussian processes for machine learning.
• An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach
• Sparse spectrum Gaussian process regression
• A unifying view of sparse approximate Gaussian process regression
• A Correspondence Between Bayesian Estimation on Stochastic Processes and Smoothing by Splines
• Title not available (Why is that?)
• Nonseparable dynamic nearest neighbor Gaussian process models for large spatio-temporal data with an application to particulate matter analysis
• On the mathematical foundations of learning
• Kernel methods in system identification, machine learning and function estimation: a survey
• Detectability and Stabilizability of Time-Varying Discrete-Time Linear Systems
• Stochastic models, estimation, and control. Vol. 1
• Title not available (Why is that?)
• The origins of kriging
• Title not available (Why is that?)
• An Approach to Statistical Spatial-Temporal Modeling of Meteorological Fields
• Dynamic models for spatiotemporal data
• Sequential Bayesian Prediction and Adaptive Sampling Algorithms for Mobile Sensor Networks
• Nonparametric identification of population models via Gaussian processes
• Multi-robots Gaussian estimation and coverage control: from client-server to peer-to-peer architectures
• Space-Time Separability in FMRI: Asymptotic Power Analysis and Cramér-Rao Lower Bounds

Cited In (3)

• An efficient implementation for spatial-temporal Gaussian process regression and its applications
• Deep networks for system identification: a survey
• Deep state-space Gaussian processes

Uses Software

• Matlab
• George

This page was built for publication: Efficient spatio-temporal Gaussian regression via Kalman filtering

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2188275)