A Sobolev-type upper bound for rates of approximation by linear combinations of Heaviside plane waves
For \(d\) odd, let \(f\) be an \( {\mathbb R}^d \to {\mathbb R}\) function, having continuous partial derivatives \(D^\alpha f(x)\) vanishing as \(| x| \to \infty\) if \(| \alpha| <d\) and satisfying, with some \(\varepsilon >0\), \(D^\alpha f(x)\| x\| ^{d+1+\varepsilon}\to 0\) if \(| \alpha| =d\). Let further \(\theta\) be the Heaviside function, \(\theta(x)=1\) if \( x \geq 0\), \(\theta (x)=0\) if \(x <0\) and let \(\Omega \subset {\mathbb R}^d\) be an arbitrary domain of unit Lebesgue measure. Then for any natural \(n\) and \(i=1, \ldots , n\) there exist real numbers \(b_i, C_i\) and \(d\)-dimensional vectors \(e_i\) such that for \(x \in {\mathbb R}^d\), \(g(x)=\sum_{i=1}^n C_i \theta (v_i \cdot x +b_i)\) one has \[ \| f -g\| _{L_2(\Omega)}\leq k_d \| f\| _{d,1,\infty}n^{-1/2}, \] where \(k_d=2^{1-d}\pi^{1-d/2}d^{d/2}/\Gamma(d/2)\), \(\| f\| _{d,1,\infty}=\max_{| \alpha| =d}\| D^\alpha f\| _{L_1({\mathbb R}^d)}.\) Applications to \(f(x)=\exp(-\| x\| ^2)\) are given.
- A better approximation for balls
- A simple lemma on greedy approximation in Hilbert space and convergence rates for projection pursuit regression and neural network training
- Approximation and learning of convex superpositions
- Comparison of worst case errors in linear and neural network approximation
- Dimension-independent bounds on the degree of approximation by neural networks
- Geometry and topology of continuous best and near best approximations
- scientific article; zbMATH DE number 5008057 (Why is no real title available?)
- scientific article; zbMATH DE number 3814037 (Why is no real title available?)
- scientific article; zbMATH DE number 3772326 (Why is no real title available?)
- scientific article; zbMATH DE number 51509 (Why is no real title available?)
- scientific article; zbMATH DE number 3482629 (Why is no real title available?)
- scientific article; zbMATH DE number 1231230 (Why is no real title available?)
- scientific article; zbMATH DE number 1182753 (Why is no real title available?)
- scientific article; zbMATH DE number 1405266 (Why is no real title available?)
- scientific article; zbMATH DE number 3071258 (Why is no real title available?)
- scientific article; zbMATH DE number 3109574 (Why is no real title available?)
- Integral combinations of Heavisides
- Real Interpolation of Sobolev Spaces on Subdomains of Rn
- The Radon transform
- Uniform approximation by neural networks
- Universal approximation bounds for superpositions of a sigmoidal function
- Complexity estimates based on integral transforms induced by computational units
- Approximation rates for neural networks with general activation functions
- Approximation of Sobolev classes by polynomials and ridge functions
- Neural network with unbounded activation functions is universal approximator
- Complexity of shallow networks representing finite mappings
- An Integral Upper Bound for Neural Network Approximation
- Minimization of Error Functionals over Perceptron Networks
- Integral combinations of Heavisides
- Can dictionary-based computational models outperform the best linear ones?
- Some comparisons of complexity in dictionary-based and linear computational models
- Integral representations of shallow neural network with Rectified Power Unit activation function
- Comparing fixed and variable-width Gaussian networks
This page was built for publication: A Sobolev-type upper bound for rates of approximation by linear combinations of Heaviside plane waves
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2643849)
