Table 1 Full-field L2 error comparison

From: Automatic network structure discovery of physics informed neural networks via knowledge distillation

Question

PINN

PINN-post

Ψ-NN

Laplace (1e−4)

11.59

4.017

0.7422

Burgers (1e−2)

14.47

3.014

1.287

Poisson (1e−2)

2.633

2.563

2.464

Flow p(1e−4)

14.89

11.78

7.838

Flow u(1e−4)

1.981

1.904

1.854

Flow v(1e−5)

1.984

1.896

1.765

  1. PINN-post refers to a model that applies hard-mapping functions as a post-processing step to the results of PINN.
Back to article page