Euclidean Fourier Neural Operators
The paper’s core claim is that standard FNOs can change the operator they represent when moved to a different periodic domain.
The authors propose EFNOs, which parameterize the spectral kernel as a continuous function of the physical wavevector instead of tying weights to integer Fourier modes. That is meant to keep the learned operator consistent across periodic domains with different shapes and sizes. They test the approach on a heat equation and on exchange-correlation potentials across crystal structures, reporting generalization to unseen grid sizes and domains. ArXiv · AI/CL/LG's note
The authors propose EFNOs, which parameterize the spectral kernel as a continuous function of the physical wavevector instead of tying weights to integer Fourier modes. That is meant to keep the learned operator consistent across periodic domains with different shapes and sizes. They test the approach on a heat equation and on exchange-correlation potentials across crystal structures, reporting generalization to unseen grid sizes and domains. ArXiv · AI/CL/LG's note
score 4