Computation complexity of deep ReLU neural networks in high-dimensional approximation

Authors Dinh Dũng, Van Kien Nguyen, Mai Xuan Thao

Publication date 1 March 2021

Abstract: The purpose of the present paper is to study the computation complexity of deep ReLU neural networks to approximate functions in H"older-Nikol'skii spaces of mixed smoothness

H_{i} n f t y^{a} l p h a (m a t h b b I^{d})

on the unit cube

m a t h b b I^{d} := [0, 1]^{d}

. In this context, for any function

f i n H_{i} n f t y^{a} l p h a (m a t h b b I^{d})

, we explicitly construct nonadaptive and adaptive deep ReLU neural networks having an output that approximates

f

with a prescribed accuracy

v a r e p s i l o n

, and prove dimension-dependent bounds for the computation complexity of this approximation, characterized by the size and the depth of this deep ReLU neural network, explicitly in

d

and

v a r e p s i l o n

. Our results show the advantage of the adaptive method of approximation by deep ReLU neural networks over nonadaptive one.

This page was built for publication: Computation complexity of deep ReLU neural networks in high-dimensional approximation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6361791)