Fig. 3
From: Physics-informed neural networks approach for 1D and 2D Gray-Scott systems
General workflow of the proposed approach. Collocation points and samples from numerical simulations are used as inputs of the pipeline. The functions u and v are returned as output of the network. Automatic differentiation is used in order to compute the loss terms related to the dynamics. The training process calibrate the network’s parameters in such a way a surrogate model is obtained. It is worth noticing that Data related components are shown for the sake of completeness since they are only applied in the two dimensional problem addressed in the following