On the Rate of Convergence of an Over-Parametrized Deep Neural Network Regression Estimate Learned by Gradient Descent
IEEE Transactions on Information Theory · 2026-03 · 72 (3) · 1777-1797
DOI: 10.1109/tit.2026.3653506 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Information Theory, 2026-03. Verfasst von Michael Kohler. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 58 Literaturangaben.
Abstract (Originalsprache)
Nonparametric regression with random design is considered. The $L_{2}$ error with integration with respect to the design measure is used as the error criterion. An over-parametrized deep neural network regression estimate with a specific sparsity structure and with logistic activation function is defined, where all weights are learned by gradient descent. It uses a network which consists of many fully connected neural networks of fixed size and depth computed in parallel and computes a linear combination of their outputs. It is shown that the estimate achieves a nearly optimal rate of convergence in case that the regression function is $(p,C)$ –smooth.
Zitieren
APA
Kohler, M. (2026). On the Rate of Convergence of an Over-Parametrized Deep Neural Network Regression Estimate Learned by Gradient Descent. IEEE Transactions on Information Theory, 72(3), 1777-1797. https://doi.org/10.1109/tit.2026.3653506
Vancouver
Kohler M. On the Rate of Convergence of an Over-Parametrized Deep Neural Network Regression Estimate Learned by Gradient Descent. IEEE Transactions on Information Theory. 2026;72(3):1777-1797. doi:10.1109/tit.2026.3653506
BibTeX
@article{Kohler2026,
author = {Kohler, Michael},
title = {On the Rate of Convergence of an Over-Parametrized Deep Neural Network Regression Estimate Learned by Gradient Descent},
journal = {IEEE Transactions on Information Theory},
year = {2026},
volume = {72},
number = {3},
pages = {1777--1797},
doi = {10.1109/tit.2026.3653506}
}
EUVIMED