Figure 2
From: Data-driven discovery of Green’s functions with human-understandable deep learning
Feature extraction from learned Green’s functions. The NNs for the learned Green’s function (A) and homogeneous solution (B) enable the extraction of qualitative and quantitative features associated with the differential operator. For example, the symmetries in the Green’s function reveal PDE invariances (C), poles of rational NNs identify singularity type and location (D), the dominant eigenvalues (E) and eigenmodes (F) of the learned Green’s function are related to the eigenvalues and eigenmodes of the differential operator.
