Spectral neural operators

DOI10.1134/S1064562423701107MaRDI QIDQ6124382zbMATH OpenFDO

Authors V. S. Fanaskov, Ivan Oseledets

Publication date 27 March 2024

Published in Doklady Mathematics (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2205.10573

zbMATH Keywords

partial differential equationsneural operators

Mathematics Subject Classification ID

Artificial neural networks and deep learning (68T07) Numerical methods for wavelets (65T60) Asymptotic approximations, asymptotic expansions (steepest descent, etc.) (41A60) Best approximation, Chebyshev systems (41A50) Finite element, Rayleigh-Ritz and Galerkin methods for boundary value problems involving PDEs (65N30) Mesh generation, refinement, and adaptive methods for boundary value problems involving PDEs (65N50) Numerical methods for discrete and fast Fourier transforms (65T50) Numerical methods for partial differential equations, boundary value problems (65N99) General integral transforms (44A05)

Abstract: A plentitude of applications in scientific computing requires the approximation of mappings between Banach spaces. Recently introduced Fourier Neural Operator (FNO) and Deep Operator Network (DeepONet) can provide this functionality. For both of these neural operators, the input function is sampled on a given grid (uniform for FNO), and the output function is parametrized by a neural network. We argue that this parametrization leads to 1) opaque output that is hard to analyze and 2) systematic bias caused by aliasing errors in the case of FNO. The alternative, advocated in this article, is to use Chebyshev and Fourier series for both domain and codomain. The resulting Spectral Neural Operator (SNO) has transparent output, never suffers from aliasing, and may include many exact (lossless) operations on functions. The functionality is based on well-developed fast, and stable algorithms from spectral methods. The implementation requires only standard numerical linear algebra. Our benchmarks show that for many operators, SNO is superior to FNO and DeepONet.

Recommendations

• Mitigating spectral bias for the multiscale operator learning with hierarchical attention
• Basis operator network: a neural network-based model for learning nonlinear operators via neural basis
• A comprehensive and fair comparison of two neural operators (with practical extensions) based on FAIR data
• Neural and spectral operator surrogates: unified construction and expression rate bounds
• Residual-based error correction for neural operator accelerated Infinite-dimensional Bayesian inverse problems

Cites work

• A numerical and theoretical study of certain nonlinear wave phenomena
• Chebyshev and Fourier spectral methods.
• Comparing Numerical Methods for Ordinary Differential Equations
• Discretization-invariant Bayesian inversion and Besov space priors
• Learning data-driven discretizations for partial differential equations
• Spectral Methods in MATLAB
• The Korteweg-de Vries two-soliton solution as interacting two single solitons.
• Wavelet neural operator for solving parametric partial differential equations in computational mechanics problems

Cited in

(7)

• Residual-based error correction for neural operator accelerated Infinite-dimensional Bayesian inverse problems
• RandONets: shallow networks with random projections for learning linear and nonlinear operators
• Mitigating spectral bias for the multiscale operator learning with hierarchical attention
• Basis operator network: a neural network-based model for learning nonlinear operators via neural basis
• Render unto numerics: orthogonal polynomial neural operator for PDEs with nonperiodic boundary conditions
• Physics-aware neural implicit solvers for multiscale, parametric PDEs with applications in heterogeneous media
• Weak neural variational inference for solving Bayesian inverse problems \textit{without} forward models: applications in elastography

This page was built for publication: Spectral neural operators

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6124382)