A deep learning solution approach for high-dimensional random differential equations

Mohammad Amin Nabian, Hadi Meidani

Research output: Contribution to journalArticle

Abstract

Developing efficient numerical algorithms for the solution of high dimensional random Partial Differential Equations (PDEs) has been a challenging task due to the well-known curse of dimensionality. We present a new solution approach for these problems based on deep learning. This approach is intrusive, entirely unsupervised, and mesh-free. Specifically, the random PDE is approximated by a feed-forward fully-connected deep residual network, with either strong or weak enforcement of initial and boundary constraints. Parameters of the approximating deep neural network are determined iteratively using variants of the Stochastic Gradient Descent (SGD) algorithm. The satisfactory accuracy of the proposed approach is numerically demonstrated on diffusion and heat conduction problems, in comparison with the converged Monte Carlo-based finite element results.

Original languageEnglish (US)
Pages (from-to)14-25
Number of pages12
JournalProbabilistic Engineering Mechanics
Volume57
DOIs
StatePublished - Jul 2019

Fingerprint

partial differential equations
learning
Partial differential equations
Differential equations
differential equations
descent
Heat conduction
conductive heat transfer
mesh
conduction
gradients
Deep learning
Deep neural networks

Keywords

  • Curse of dimensionality
  • Deep learning
  • Deep neural networks
  • Least squares
  • Random differential equations
  • Residual networks

ASJC Scopus subject areas

  • Statistical and Nonlinear Physics
  • Civil and Structural Engineering
  • Nuclear Energy and Engineering
  • Condensed Matter Physics
  • Aerospace Engineering
  • Ocean Engineering
  • Mechanical Engineering

Cite this

A deep learning solution approach for high-dimensional random differential equations. / Nabian, Mohammad Amin; Meidani, Hadi.

In: Probabilistic Engineering Mechanics, Vol. 57, 07.2019, p. 14-25.

Research output: Contribution to journalArticle

@article{2be44d87929c4f48b59b58a71667893c,
title = "A deep learning solution approach for high-dimensional random differential equations",
abstract = "Developing efficient numerical algorithms for the solution of high dimensional random Partial Differential Equations (PDEs) has been a challenging task due to the well-known curse of dimensionality. We present a new solution approach for these problems based on deep learning. This approach is intrusive, entirely unsupervised, and mesh-free. Specifically, the random PDE is approximated by a feed-forward fully-connected deep residual network, with either strong or weak enforcement of initial and boundary constraints. Parameters of the approximating deep neural network are determined iteratively using variants of the Stochastic Gradient Descent (SGD) algorithm. The satisfactory accuracy of the proposed approach is numerically demonstrated on diffusion and heat conduction problems, in comparison with the converged Monte Carlo-based finite element results.",
keywords = "Curse of dimensionality, Deep learning, Deep neural networks, Least squares, Random differential equations, Residual networks",
author = "Nabian, {Mohammad Amin} and Hadi Meidani",
year = "2019",
month = "7",
doi = "10.1016/j.probengmech.2019.05.001",
language = "English (US)",
volume = "57",
pages = "14--25",
journal = "Probabilistic Engineering Mechanics",
issn = "0266-8920",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - A deep learning solution approach for high-dimensional random differential equations

AU - Nabian, Mohammad Amin

AU - Meidani, Hadi

PY - 2019/7

Y1 - 2019/7

N2 - Developing efficient numerical algorithms for the solution of high dimensional random Partial Differential Equations (PDEs) has been a challenging task due to the well-known curse of dimensionality. We present a new solution approach for these problems based on deep learning. This approach is intrusive, entirely unsupervised, and mesh-free. Specifically, the random PDE is approximated by a feed-forward fully-connected deep residual network, with either strong or weak enforcement of initial and boundary constraints. Parameters of the approximating deep neural network are determined iteratively using variants of the Stochastic Gradient Descent (SGD) algorithm. The satisfactory accuracy of the proposed approach is numerically demonstrated on diffusion and heat conduction problems, in comparison with the converged Monte Carlo-based finite element results.

AB - Developing efficient numerical algorithms for the solution of high dimensional random Partial Differential Equations (PDEs) has been a challenging task due to the well-known curse of dimensionality. We present a new solution approach for these problems based on deep learning. This approach is intrusive, entirely unsupervised, and mesh-free. Specifically, the random PDE is approximated by a feed-forward fully-connected deep residual network, with either strong or weak enforcement of initial and boundary constraints. Parameters of the approximating deep neural network are determined iteratively using variants of the Stochastic Gradient Descent (SGD) algorithm. The satisfactory accuracy of the proposed approach is numerically demonstrated on diffusion and heat conduction problems, in comparison with the converged Monte Carlo-based finite element results.

KW - Curse of dimensionality

KW - Deep learning

KW - Deep neural networks

KW - Least squares

KW - Random differential equations

KW - Residual networks

UR - http://www.scopus.com/inward/record.url?scp=85065730866&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065730866&partnerID=8YFLogxK

U2 - 10.1016/j.probengmech.2019.05.001

DO - 10.1016/j.probengmech.2019.05.001

M3 - Article

AN - SCOPUS:85065730866

VL - 57

SP - 14

EP - 25

JO - Probabilistic Engineering Mechanics

JF - Probabilistic Engineering Mechanics

SN - 0266-8920

ER -