Understanding and Mitigating Extrapolation Failures in Physics-Informed Neural Networks
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:LSUUMCDBrecord.jsonopen to challenge →
read the original abstract
Physics-informed Neural Networks (PINNs) have recently gained popularity due to their effective approximation of partial differential equations (PDEs) using deep neural networks (DNNs). However, their out of domain behavior is not well understood, with previous work speculating that the presence of high frequency components in the solution function might be to blame for poor extrapolation performance. In this paper, we study the extrapolation behavior of PINNs on a representative set of PDEs of different types, including high-dimensional PDEs. We find that failure to extrapolate is not caused by high frequencies in the solution function, but rather by shifts in the support of the Fourier spectrum over time. We term these spectral shifts and quantify them by introducing a Weighted Wasserstein-Fourier distance (WWF). We show that the WWF can be used to predict PINN extrapolation performance, and that in the absence of significant spectral shifts, PINN predictions stay close to the true solution even in extrapolation. Finally, we propose a transfer learning-based strategy to mitigate the effects of larger spectral shifts, which decreases extrapolation errors by up to 82%.
This paper has not been read by Pith yet.
Forward citations
Cited by 3 Pith papers
-
SPLIT-PINN: Separable Probability Learning Technique via Physics-Informed Neural Networks for High-Dimensional Probabilistic Modeling
SPLIT-PINN infers drift fields in Liouville transport equations from data using marginal corrections and orthogonality constraints to enable probabilistic predictions of microstructural evolution across polycrystal re...
-
Accelerating 4D Hyperspectral Imaging through Physics-Informed Neural Representation and Adaptive Sampling
A physics-informed MLP reconstructs high-fidelity 4D spectra from only 1/32 of the samples in experimental 2DIR hyperspectral imaging.
-
MoGERNN: An Inductive Traffic Predictor for Unobserved Locations
MoGERNN uses a mixture-of-graph-experts module and encoder-decoder structure to predict traffic states at unobserved locations and remain effective when the sensor network changes.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.