English

Analytic function approximation by path norm regularized deep networks

Machine Learning 2022-11-18 v4 Machine Learning

Abstract

We show that neural networks with absolute value activation function and with the path norm, the depth, the width and the network weights having logarithmic dependence on 1/ε1/\varepsilon can ε\varepsilon-approximate functions that are analytic on certain regions of Cd\mathbb{C}^d.

Keywords

Cite

@article{arxiv.2104.02095,
  title  = {Analytic function approximation by path norm regularized deep networks},
  author = {Aleksandr Beknazaryan},
  journal= {arXiv preprint arXiv:2104.02095},
  year   = {2022}
}
R2 v1 2026-06-24T00:51:56.202Z