Replacing Automatic Differentiation by Sobolev Cubatures fastens Physics Informed Neural Nets and strengthens their Approximation Power
This addresses a computational bottleneck for researchers and practitioners using PINNs to solve forward and inverse PDE problems, though it appears incremental as it builds on existing PINN frameworks.
The authors tackled the computational inefficiency of Physics-Informed Neural Networks (PINNs) by replacing automatic differentiation with Sobolev cubatures, resulting in one-to-two orders of magnitude speed-up and closer solution approximations for PDE problems.
We present a novel class of approximations for variational losses, being applicable for the training of physics-informed neural nets (PINNs). The loss formulation reflects classic Sobolev space theory for partial differential equations and their weak formulations. The loss computation rests on an extension of Gauss-Legendre cubatures, we term Sobolev cubatures, replacing automatic differentiation (A.D.). We prove the runtime complexity of training the resulting Soblev-PINNs (SC-PINNs) to be less than required by PINNs relying on A.D. On top of one-to-two order of magnitude speed-up the SC-PINNs are demonstrated to achieve closer solution approximations for prominent forward and inverse PDE problems than established PINNs achieve.