Fig. 5: Burgers case results.
From: Automatic network structure discovery of physics informed neural networks via knowledge distillation
a The evolution of parameters under Ψ-NN method. The loss curve here serves as an indicator of residual stability rather than final accuracy. b Cluster centers of Ψ-NN in Laplace equation. The x-axis represents the absolute value of the weights, and the y-values are given randomly in order to better visualize the distribution. Negative values are shown as red dots, and positive values are shown as blue dots. The cluster distance is set to 0.1. The right column contains distributions of biases, left column contains distributions of weights. The first to fourth rows correspond to the clustering results of the network parameters for the first to third hidden layers and the output layer, respectively. c The structure of student NN after parameter replacement in Laplace equation.
