Spectral Barron space for deep neural network approximation

2023/09/02 by Liao, Yulei, Ming, Pingbing · 5 citations
#32C22 #32K05 #33C20 #41A25 #41A46 #42A38 #68T07 #FOS: Mathematics #Numerical Analysis (math.NA)

paper · doi:10.48550/arxiv.2309.00788

Abstract

We prove the sharp embedding between the spectral Barron space and the Besov space with embedding constants independent of the input dimension. Given the spectral Barron space as the target function space, we prove a dimension-free convergence result that if the neural network contains L hidden layers with N units per layer, then the upper and lower bounds of the L²-approximation error are O(N^-sL) with 0 < sL≤ 1/2, where s≥ 0 is the smoothness index of the spectral Barron space.

Cited by

• Time-Frequency Analysis for Neural Networks
• Spectral Barron spaces of vector-valued functions on compact groups
• Statistical Learning Guarantees for Group-Invariant Barron Functions
• Neural Network Approximation of Solutions to Fractional Parabolic Partial Differential Equations
• Barron regularity of many particle Schrödinger eigenfunctions

Related

• Sharp uniform approximation for spectral Barron functions by deep neural networks
• Sobolev-type embeddings for neural network approximation spaces
• Space-Time Approximation with Shallow Neural Networks in Fourier Lebesgue spaces
• Weighted Sobolev Approximation Rates for Neural Networks on Unbounded Domains
• Approximation spaces of deep neural networks