Spatiotemporal Pattern Extraction by Spectral Analysis of Vector-Valued Observables

Dimitrios Giannakis, Abbas Ourmazd, Joanna Slawinska, Zhizhen Zhao

Research output: Contribution to journalArticle

Abstract

We present a data-driven framework for extracting complex spatiotemporal patterns generated by ergodic dynamical systems. Our approach, called vector-valued spectral analysis (VSA), is based on an eigendecomposition of a kernel integral operator acting on a Hilbert space of vector-valued observables of the system, taking values in a space of functions (scalar fields) on a spatial domain. This operator is constructed by combining aspects of the theory of operator-valued kernels for multitask machine learning with delay-coordinate maps of dynamical systems. In contrast to conventional eigendecomposition techniques, which decompose the input data into pairs of temporal and spatial modes with a separable, tensor product structure, the patterns recovered by VSA can be manifestly non-separable, requiring only a modest number of modes to represent signals with intermittency in both space and time. Moreover, the kernel construction naturally quotients out dynamical symmetries in the data and exhibits an asymptotic commutativity property with the Koopman evolution operator of the system, enabling decomposition of multiscale signals into dynamically intrinsic patterns. Application of VSA to the Kuramoto–Sivashinsky model demonstrates significant performance gains in efficient and meaningful decomposition over eigendecomposition techniques utilizing scalar-valued kernels.

Original languageEnglish (US)
JournalJournal of Nonlinear Science
DOIs
StatePublished - Jan 1 2019

Fingerprint

Spatio-temporal Patterns
Spectral Analysis
Spectrum analysis
kernel
Decompose
Dynamical systems
Dynamical system
Multi-task Learning
Kernel Operator
Decomposition
Evolution Operator
Nonseparable
Commutativity
Intermittency
Hilbert spaces
Function Fields
Operator
Data-driven
Integral Operator
Tensor Product

Keywords

  • Dynamical symmetries
  • Dynamical systems
  • Kernel methods
  • Koopman operators
  • Spatiotemporal patterns
  • Spectral decomposition

ASJC Scopus subject areas

  • Modeling and Simulation
  • Engineering(all)
  • Applied Mathematics

Cite this

Spatiotemporal Pattern Extraction by Spectral Analysis of Vector-Valued Observables. / Giannakis, Dimitrios; Ourmazd, Abbas; Slawinska, Joanna; Zhao, Zhizhen.

In: Journal of Nonlinear Science, 01.01.2019.

Research output: Contribution to journalArticle

@article{3bba41def4d14ca9a7f34cf5e0600204,
title = "Spatiotemporal Pattern Extraction by Spectral Analysis of Vector-Valued Observables",
abstract = "We present a data-driven framework for extracting complex spatiotemporal patterns generated by ergodic dynamical systems. Our approach, called vector-valued spectral analysis (VSA), is based on an eigendecomposition of a kernel integral operator acting on a Hilbert space of vector-valued observables of the system, taking values in a space of functions (scalar fields) on a spatial domain. This operator is constructed by combining aspects of the theory of operator-valued kernels for multitask machine learning with delay-coordinate maps of dynamical systems. In contrast to conventional eigendecomposition techniques, which decompose the input data into pairs of temporal and spatial modes with a separable, tensor product structure, the patterns recovered by VSA can be manifestly non-separable, requiring only a modest number of modes to represent signals with intermittency in both space and time. Moreover, the kernel construction naturally quotients out dynamical symmetries in the data and exhibits an asymptotic commutativity property with the Koopman evolution operator of the system, enabling decomposition of multiscale signals into dynamically intrinsic patterns. Application of VSA to the Kuramoto–Sivashinsky model demonstrates significant performance gains in efficient and meaningful decomposition over eigendecomposition techniques utilizing scalar-valued kernels.",
keywords = "Dynamical symmetries, Dynamical systems, Kernel methods, Koopman operators, Spatiotemporal patterns, Spectral decomposition",
author = "Dimitrios Giannakis and Abbas Ourmazd and Joanna Slawinska and Zhizhen Zhao",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/s00332-019-09548-1",
language = "English (US)",
journal = "Journal of Nonlinear Science",
issn = "0938-8974",
publisher = "Springer New York",

}

TY - JOUR

T1 - Spatiotemporal Pattern Extraction by Spectral Analysis of Vector-Valued Observables

AU - Giannakis, Dimitrios

AU - Ourmazd, Abbas

AU - Slawinska, Joanna

AU - Zhao, Zhizhen

PY - 2019/1/1

Y1 - 2019/1/1

N2 - We present a data-driven framework for extracting complex spatiotemporal patterns generated by ergodic dynamical systems. Our approach, called vector-valued spectral analysis (VSA), is based on an eigendecomposition of a kernel integral operator acting on a Hilbert space of vector-valued observables of the system, taking values in a space of functions (scalar fields) on a spatial domain. This operator is constructed by combining aspects of the theory of operator-valued kernels for multitask machine learning with delay-coordinate maps of dynamical systems. In contrast to conventional eigendecomposition techniques, which decompose the input data into pairs of temporal and spatial modes with a separable, tensor product structure, the patterns recovered by VSA can be manifestly non-separable, requiring only a modest number of modes to represent signals with intermittency in both space and time. Moreover, the kernel construction naturally quotients out dynamical symmetries in the data and exhibits an asymptotic commutativity property with the Koopman evolution operator of the system, enabling decomposition of multiscale signals into dynamically intrinsic patterns. Application of VSA to the Kuramoto–Sivashinsky model demonstrates significant performance gains in efficient and meaningful decomposition over eigendecomposition techniques utilizing scalar-valued kernels.

AB - We present a data-driven framework for extracting complex spatiotemporal patterns generated by ergodic dynamical systems. Our approach, called vector-valued spectral analysis (VSA), is based on an eigendecomposition of a kernel integral operator acting on a Hilbert space of vector-valued observables of the system, taking values in a space of functions (scalar fields) on a spatial domain. This operator is constructed by combining aspects of the theory of operator-valued kernels for multitask machine learning with delay-coordinate maps of dynamical systems. In contrast to conventional eigendecomposition techniques, which decompose the input data into pairs of temporal and spatial modes with a separable, tensor product structure, the patterns recovered by VSA can be manifestly non-separable, requiring only a modest number of modes to represent signals with intermittency in both space and time. Moreover, the kernel construction naturally quotients out dynamical symmetries in the data and exhibits an asymptotic commutativity property with the Koopman evolution operator of the system, enabling decomposition of multiscale signals into dynamically intrinsic patterns. Application of VSA to the Kuramoto–Sivashinsky model demonstrates significant performance gains in efficient and meaningful decomposition over eigendecomposition techniques utilizing scalar-valued kernels.

KW - Dynamical symmetries

KW - Dynamical systems

KW - Kernel methods

KW - Koopman operators

KW - Spatiotemporal patterns

KW - Spectral decomposition

UR - http://www.scopus.com/inward/record.url?scp=85065716938&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065716938&partnerID=8YFLogxK

U2 - 10.1007/s00332-019-09548-1

DO - 10.1007/s00332-019-09548-1

M3 - Article

JO - Journal of Nonlinear Science

JF - Journal of Nonlinear Science

SN - 0938-8974

ER -