Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
CAFE+FNO enhances Fourier Neural Operators by incorporating Content-Aware Frequency Encoding+ (CAFE+) to address frequency truncation limitations. CAFE+ constructs frequency interactions through multiplicative composition of Fourier-Chebyshev features, enabling richer representation of high-frequency variations. Experiments on five PDE benchmarks demonstrate CAFE+FNO's superior performance compared to existing FNO variants.
Multiplicative composition of Fourier-Chebyshev features unlocks richer PDE solution representations than implicit MLP composition.
The Fourier Neural Operator (FNO) learns solution operators of partial differential equations (PDEs) through Fourier-space kernel parameterization, but frequency truncation can limit the learning of high-frequency variations. AM-FNO and SirenFNO generate kernels for all grid modes from spectral coordinates using shared networks, making coordinate encoding and generator design important. Recent work on implicit neural representations (INRs) has proposed constructing frequency interactions through explicit feature composition rather than relying on subsequent MLPs to form them implicitly. Building on this approach, we propose CAFE+FNO, which incorporates Content-Aware Frequency Encoding+ (CAFE+) into Fourier kernel generation. CAFE+ combines Fourier--Chebyshev features through parallel affine branches and a Hadamard product, forming interactions within and across the two feature families. A kernel MLP maps the resulting representation of each normalized spectral coordinate to a complex channel-mixing matrix. Each layer shares its generator across all stored modes, making the number of trainable parameters independent of the number of modes for a fixed architecture. We compare CAFE+FNO with existing FNO variants on five PDE benchmarks and conduct ablation studies on basis configuration, multiplicative composition, and bandwidth learnability. Code and experimental configurations are available at https://github.com/fabsk101/CAFEPlusFNO.git.