Less Emphasis on Difficult Layer Regions: Curriculum Learning for Singularly Perturbed Convection-Diffusion-Reaction Problems

10/23/2022
by   Yufeng Wang, et al.
0

Although Physics-Informed Neural Networks (PINNs) have been successfully applied to various differential equations, accurately solving perturbed convection-diffusion-reaction problems is still extremely challenging for PINNs. This paper investigates the source of the learning difficulties and finds that the rapid transition of potential solution in the layer region causes the failure of convergence. Based on this finding, we present a curriculum learning method that encourages neural networks to “prioritize the learning on easier non-layer regions”. The method helps PINNs to dynamically adjust the training data weights, speed up the learning procedure, and ultimately significantly improve the accuracy of the network approximation. Extensive evaluation on multiple typical model equations shows that the proposed approach accurately captures the resolution of the layer regions, and achieves multiple orders of magnitude lower root-mean-squared error than ordinary PINNs. We provide our PyTorch code at https://github.com/WYu-Feng/CLPINN

READ FULL TEXT

page 13

page 14

page 15

page 16

research
07/21/2020

Unsupervised Learning of Solutions to Differential Equations with Generative Adversarial Networks

Solutions to differential equations are of significant scientific and en...
research
08/19/2022

Semi-analytic PINN methods for singularly perturbed boundary value problems

We propose a new semi-analytic physics informed neural network (PINN) to...
research
11/09/2020

Stiff-PINN: Physics-Informed Neural Network for Stiff Chemical Kinetics

Recently developed physics-informed neural network (PINN) has achieved s...
research
11/03/2021

Schwarz Waveform Relaxation Physics-Informed Neural Networks for Solving Advection-Diffusion-Reaction Equations

This paper develops a physics-informed neural network (PINN) based on th...
research
09/02/2021

Characterizing possible failure modes in physics-informed neural networks

Recent work in scientific machine learning has developed so-called physi...
research
11/01/2021

Numerical Approximation in CFD Problems Using Physics Informed Machine Learning

The thesis focuses on various techniques to find an alternate approximat...

Please sign up or login with your details

Forgot password? Click here to reset