Machine learning with observers predicts complex spatiotemporal behavior

George Neofotistos, Marios Mattheakis, Georgios D. Barmparis, Johanne Hizanidis, Georgios Tsironis, Efthimios Kaxiras

Research output: Contribution to journalArticle

Abstract

Chimeras and branching are two archetypical complex phenomena that appear in many physical systems; because of their different intrinsic dynamics, they delineate opposite non-trivial limits in the complexity of wave motion and present severe challenges in predicting chaotic and singular behavior in extended physical systems. We report on the long-term forecasting capability of Long Short-Term Memory (LSTM) and reservoir computing (RC) recurrent neural networks, when they are applied to the spatiotemporal evolution of turbulent chimeras in simulated arrays of coupled superconducting quantum interference devices (SQUIDs) or lasers, and branching in the electronic flow of two-dimensional graphene with random potential. We propose a new method in which we assign one LSTM network to each system node except for "observer" nodes which provide continual "ground truth" measurements as input; we refer to this method as "Observer LSTM" (OLSTM). We demonstrate that even a small number of observers greatly improves the data-driven (model-free) long-term forecasting capability of the LSTM networks and provide the framework for a consistent comparison between the RC and LSTM methods. We find that RC requires smaller training datasets than OLSTMs, but the latter require fewer observers. Both methods are benchmarked against Feed-Forward neural networks (FNNs), also trained to make predictions with observers (OFNNs).

Original languageEnglish
Article number24
JournalFrontiers in Physics
Volume7
Issue numberMAR
DOIs
Publication statusPublished - Jan 1 2019

Fingerprint

machine learning
Long-Term Memory
Memory Term
Short-Term Memory
Learning systems
Observer
Machine Learning
Predict
forecasting
Forecasting
Branching
Computing
Quantum Interference
Random Potential
ground truth
Graphite
Recurrent neural networks
Feedforward neural networks
Graphene
SQUIDs

Keywords

  • Branched flow
  • Chimera state
  • Graphene
  • Long short-term memory
  • Machine learning
  • Prediction
  • Reservoir computing

ASJC Scopus subject areas

  • Biophysics
  • Materials Science (miscellaneous)
  • Mathematical Physics
  • Physics and Astronomy(all)
  • Physical and Theoretical Chemistry

Cite this

Neofotistos, G., Mattheakis, M., Barmparis, G. D., Hizanidis, J., Tsironis, G., & Kaxiras, E. (2019). Machine learning with observers predicts complex spatiotemporal behavior. Frontiers in Physics, 7(MAR), [24]. https://doi.org/10.3389/fphy.2019.00024

Machine learning with observers predicts complex spatiotemporal behavior. / Neofotistos, George; Mattheakis, Marios; Barmparis, Georgios D.; Hizanidis, Johanne; Tsironis, Georgios; Kaxiras, Efthimios.

In: Frontiers in Physics, Vol. 7, No. MAR, 24, 01.01.2019.

Research output: Contribution to journalArticle

Neofotistos, G, Mattheakis, M, Barmparis, GD, Hizanidis, J, Tsironis, G & Kaxiras, E 2019, 'Machine learning with observers predicts complex spatiotemporal behavior', Frontiers in Physics, vol. 7, no. MAR, 24. https://doi.org/10.3389/fphy.2019.00024
Neofotistos G, Mattheakis M, Barmparis GD, Hizanidis J, Tsironis G, Kaxiras E. Machine learning with observers predicts complex spatiotemporal behavior. Frontiers in Physics. 2019 Jan 1;7(MAR). 24. https://doi.org/10.3389/fphy.2019.00024
Neofotistos, George ; Mattheakis, Marios ; Barmparis, Georgios D. ; Hizanidis, Johanne ; Tsironis, Georgios ; Kaxiras, Efthimios. / Machine learning with observers predicts complex spatiotemporal behavior. In: Frontiers in Physics. 2019 ; Vol. 7, No. MAR.
@article{f8c558365720484790aaaff78af04ed1,
title = "Machine learning with observers predicts complex spatiotemporal behavior",
abstract = "Chimeras and branching are two archetypical complex phenomena that appear in many physical systems; because of their different intrinsic dynamics, they delineate opposite non-trivial limits in the complexity of wave motion and present severe challenges in predicting chaotic and singular behavior in extended physical systems. We report on the long-term forecasting capability of Long Short-Term Memory (LSTM) and reservoir computing (RC) recurrent neural networks, when they are applied to the spatiotemporal evolution of turbulent chimeras in simulated arrays of coupled superconducting quantum interference devices (SQUIDs) or lasers, and branching in the electronic flow of two-dimensional graphene with random potential. We propose a new method in which we assign one LSTM network to each system node except for {"}observer{"} nodes which provide continual {"}ground truth{"} measurements as input; we refer to this method as {"}Observer LSTM{"} (OLSTM). We demonstrate that even a small number of observers greatly improves the data-driven (model-free) long-term forecasting capability of the LSTM networks and provide the framework for a consistent comparison between the RC and LSTM methods. We find that RC requires smaller training datasets than OLSTMs, but the latter require fewer observers. Both methods are benchmarked against Feed-Forward neural networks (FNNs), also trained to make predictions with observers (OFNNs).",
keywords = "Branched flow, Chimera state, Graphene, Long short-term memory, Machine learning, Prediction, Reservoir computing",
author = "George Neofotistos and Marios Mattheakis and Barmparis, {Georgios D.} and Johanne Hizanidis and Georgios Tsironis and Efthimios Kaxiras",
year = "2019",
month = "1",
day = "1",
doi = "10.3389/fphy.2019.00024",
language = "English",
volume = "7",
journal = "Frontiers in Physics",
issn = "2296-424X",
publisher = "Frontiers Media S. A.",
number = "MAR",

}

TY - JOUR

T1 - Machine learning with observers predicts complex spatiotemporal behavior

AU - Neofotistos, George

AU - Mattheakis, Marios

AU - Barmparis, Georgios D.

AU - Hizanidis, Johanne

AU - Tsironis, Georgios

AU - Kaxiras, Efthimios

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Chimeras and branching are two archetypical complex phenomena that appear in many physical systems; because of their different intrinsic dynamics, they delineate opposite non-trivial limits in the complexity of wave motion and present severe challenges in predicting chaotic and singular behavior in extended physical systems. We report on the long-term forecasting capability of Long Short-Term Memory (LSTM) and reservoir computing (RC) recurrent neural networks, when they are applied to the spatiotemporal evolution of turbulent chimeras in simulated arrays of coupled superconducting quantum interference devices (SQUIDs) or lasers, and branching in the electronic flow of two-dimensional graphene with random potential. We propose a new method in which we assign one LSTM network to each system node except for "observer" nodes which provide continual "ground truth" measurements as input; we refer to this method as "Observer LSTM" (OLSTM). We demonstrate that even a small number of observers greatly improves the data-driven (model-free) long-term forecasting capability of the LSTM networks and provide the framework for a consistent comparison between the RC and LSTM methods. We find that RC requires smaller training datasets than OLSTMs, but the latter require fewer observers. Both methods are benchmarked against Feed-Forward neural networks (FNNs), also trained to make predictions with observers (OFNNs).

AB - Chimeras and branching are two archetypical complex phenomena that appear in many physical systems; because of their different intrinsic dynamics, they delineate opposite non-trivial limits in the complexity of wave motion and present severe challenges in predicting chaotic and singular behavior in extended physical systems. We report on the long-term forecasting capability of Long Short-Term Memory (LSTM) and reservoir computing (RC) recurrent neural networks, when they are applied to the spatiotemporal evolution of turbulent chimeras in simulated arrays of coupled superconducting quantum interference devices (SQUIDs) or lasers, and branching in the electronic flow of two-dimensional graphene with random potential. We propose a new method in which we assign one LSTM network to each system node except for "observer" nodes which provide continual "ground truth" measurements as input; we refer to this method as "Observer LSTM" (OLSTM). We demonstrate that even a small number of observers greatly improves the data-driven (model-free) long-term forecasting capability of the LSTM networks and provide the framework for a consistent comparison between the RC and LSTM methods. We find that RC requires smaller training datasets than OLSTMs, but the latter require fewer observers. Both methods are benchmarked against Feed-Forward neural networks (FNNs), also trained to make predictions with observers (OFNNs).

KW - Branched flow

KW - Chimera state

KW - Graphene

KW - Long short-term memory

KW - Machine learning

KW - Prediction

KW - Reservoir computing

UR - http://www.scopus.com/inward/record.url?scp=85065321878&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065321878&partnerID=8YFLogxK

U2 - 10.3389/fphy.2019.00024

DO - 10.3389/fphy.2019.00024

M3 - Article

AN - SCOPUS:85065321878

VL - 7

JO - Frontiers in Physics

JF - Frontiers in Physics

SN - 2296-424X

IS - MAR

M1 - 24

ER -