Deep ReLU neural networks in high-dimensional approximation
| Autor: | Van Kien Nguyen, Dinh Dũng |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Smoothness (probability theory)
Artificial neural network Cognitive Neuroscience Numerical Analysis (math.NA) Function (mathematics) Upper and lower bounds Sobolev space Dimension (vector space) Computer Systems Artificial Intelligence Unit cube Approximation error FOS: Mathematics Applied mathematics Neural Networks Computer Mathematics - Numerical Analysis Mathematics |
| Zdroj: | Neural Networks. 142:619-635 |
| ISSN: | 0893-6080 |
| DOI: | 10.1016/j.neunet.2021.07.027 |
| Popis: | We study the computation complexity of deep ReLU (Rectified Linear Unit) neural networks for the approximation of functions from the H\"older-Zygmund space of mixed smoothness defined on the $d$-dimensional unit cube when the dimension $d$ may be very large. The approximation error is measured in the norm of isotropic Sobolev space. For every function $f$ from the H\"older-Zygmund space of mixed smoothness, we explicitly construct a deep ReLU neural network having an output that approximates $f$ with a prescribed accuracy $\varepsilon$, and prove tight dimension-dependent upper and lower bounds of the computation complexity of this approximation, characterized as the size and the depth of this deep ReLU neural network, explicitly in $d$ and $\varepsilon$. The proof of these results are in particular, relied on the approximation by sparse-grid sampling recovery based on the Faber series. Comment: 5 figures |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect