Variational training of neural network approximations of solution maps for physical models

Abstract

A novel solve-training framework is proposed to train neural network in representing low dimensional solution maps of physical models. Solve-training framework uses the neural network as the ansatz of the solution map and train the network variationally via loss functions from the underlying physical models. Solve-training framework avoids expensive data preparation in the traditional supervised training procedure, which prepares labels for input data, and still achieves effective representation of the solution map adapted to the input data distribution. The efficiency of solve-training framework is demonstrated through obtaining solutions maps for linear and nonlinear elliptic equations, and maps from potentials to ground states of linear and nonlinear Schrödinger equations.

References in corpus (16)

• Artificial Neural Networks for Solving Ordinary and Partial Differential Equations
• DGM: A deep learning algorithm for solving partial differential equations
• Solving the Quantum Many-Body Problem with Artificial Neural Networks
• Solving high-dimensional partial differential equations using deep learning
• Physics-Constrained Deep Learning for High-dimensional Surrogate Modeling and Uncertainty Quantification without Labeled Data
• A unified deep artificial neural network approach to partial differential equations in complex geometries
• PDE-Net: Learning PDEs from Data
• BCR-Net: a neural network based on the nonstandard wavelet form
• Learning Neural PDE Solvers with Convergence Guarantees
• Learning to Optimize Multigrid PDE Solvers
• Unsupervised Deep Learning Algorithm for PDE-based Forward and Inverse Problems
• A multiscale neural network based on hierarchical matrices
• Butterfly-Net: Optimal Function Representation Based on Convolutional Neural Networks
• Study on a Poisson's Equation Solver Based On Deep Learning Technique
• Deep Multigrid: learning prolongation and restriction matrices
• SwitchNet: a neural network model for forward and inverse scattering problems

Cited by in corpus (16)

• The Random Feature Model for Input-Output Maps between Banach Spaces
• Solving Electrical Impedance Tomography with Deep Learning
• Using neural networks to accelerate the solution of the Boltzmann equation
• Meta-learning Pseudo-differential Operators with Deep Neural Networks
• A new efficient approximation scheme for solving high-dimensional semilinear PDEs: control variate method for Deep BSDE solver
• Operator Learning Using Random Features: A Tool for Scientific Computing
• Butterfly-Net: Optimal Function Representation Based on Convolutional Neural Networks
• Solving Optical Tomography with Deep Learning
• Cell-average based neural network method for hyperbolic and parabolic partial differential equations
• Meta-MgNet: Meta Multigrid Networks for Solving Parameterized Partial Differential Equations
• Predicting continuum breakdown with deep neural networks
• Learning to Discretize: Solving 1D Scalar Conservation Laws via Deep Reinforcement Learning
• Forward and backward problems for coupled subdiffusion systems
• Butterfly-Net2: Simplified Butterfly-Net and Fourier Transform Initialization
• A reduced order Schwarz method for nonlinear multiscale elliptic equations based on two-layer neural networks
• Accuracy and Architecture Studies of Residual Neural Network solving Ordinary Differential Equations