An Integral Upper Bound for Neural Network Approximation

被引：32

作者：

Kainen, Paul C. ^{[1
]}

Kurkova, Vera ^{[2
]}

机构：

[1] Georgetown Univ, Dept Math, Washington, DC 20057 USA

[2] Acad Sci Czech Republic, Inst Comp Sci, Prague, Czech Republic

来源：

NEURAL COMPUTATION | 2009年 / 21卷 / 10期

关键词：

SUPERPOSITIONS; RATES; SPACES;

D O I：

10.1162/neco.2009.04-08-745

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Complexity of one-hidden-layer networks is studied using tools from nonlinear approximation and integration theory. For functions with suitable integral representations in the form of networks with infinitely many hidden units, upper bounds are derived on the speed of decrease of approximation error as the number of network units increases. These bounds are obtained for various norms using the framework of Bochner integration. Results are applied to perceptron networks.

引用

页码：2970 / 2989

页数：20

共 22 条

[1]

Adams RA., 2003, SOBOLEV SPACES, V2

[2]

[Anonymous], T AM MATH SOC, DOI DOI 10.2307/1990404

[3]

Barron, 1992, P 7 YAL WORKSH AD LE, P69

[4] UNIVERSAL APPROXIMATION BOUNDS FOR SUPERPOSITIONS OF A SIGMOIDAL FUNCTION [J].

BARRON, AR .

IEEE TRANSACTIONS ON INFORMATION THEORY, 1993, 39 (03) :930-945

[5]

Darken C., 1993, Proceeding of the Sixth Annual ACM Conference on Computational Learning Theory, P303, DOI 10.1145/168304.168357

[6]

Donahue MJ, 1997, CONSTR APPROX, V13, P187

[7]

Girosi F., 1993, ARTIFICIAL NEURAL NE, P97

[8] Approximation and learning of convex superpositions [J].

Gurvits, L ;

Koiran, P .

JOURNAL OF COMPUTER AND SYSTEM SCIENCES, 1997, 55 (01) :161-170

[9]

Hewitt E., 1965, Real and Abstract Analysis

[10] REPRESENTATION OF FUNCTIONS BY SUPERPOSITIONS OF A STEP OR SIGMOID FUNCTION AND THEIR APPLICATIONS TO NEURAL NETWORK THEORY [J].

ITO, Y .

NEURAL NETWORKS, 1991, 4 (03) :385-394

← 1 2 3 →