SirenFNO: Efficient and Full Frequency Learning of Fourier Neural Operators

In three linesSirenFNO combines Fourier neural operators (FNO) with sinusoidal representation networks (SIREN) to learn full-spectrum without frequency truncation. The framework reduces parameters by 4–15× on PDE benchmarks while eliminating spectral bias toward low frequencies; functional tensor decomposition variants achieve 73× parameter reduction.

Summary generated by Claude — human-verified