Lower estimation of approximation rate for neural networks

被引:2
作者
Cao FeiLong [1 ]
Zhang YongQuan [2 ]
Xu ZongBen [2 ]
机构
[1] China Jiliang Univ, Coll Sci, Hangzhou 310018, Peoples R China
[2] Xi An Jiao Tong Univ, Inst Informat & Syst Sci, Xian 710049, Peoples R China
来源
SCIENCE IN CHINA SERIES F-INFORMATION SCIENCES | 2009年 / 52卷 / 08期
基金
中国国家自然科学基金;
关键词
feedforward neural networks; approximation; topology structure of hidden layer; rate; lower; MULTILAYER FEEDFORWARD NETWORKS; UNIVERSAL APPROXIMATION; ESSENTIAL ORDER; BOUNDS; OPERATORS;
D O I
10.1007/s11432-009-0027-7
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Let SFd and Pi(phi,n,d) = {Sigma(n)(j=1)b(j)phi(omega j.x+theta j) :b(j),theta(j)is an element of R,omega(j)is an element of R-d} be the set of periodic and Lebesgue's square-integrable functions and the set of feedforward neural network (FNN) functions, respectively. Denote by dist (SFd, Pi(phi,n,d)) the deviation of the set SFd from the set pi(phi,n,d). A main purpose of this paper is to estimate the deviation. In particular, based on the Fourier transforms and the theory of approximation, a lower estimation for dist (SFd, Pi(phi,n,d)) is proved. That is, dist(SFd,Pi(phi,n,d)) >= C/(nlog(2)(n))(1/2). The obtained estimation depends only on the number of neuron in the hidden layer, and is independent of the approximated target functions and dimensional number of input. This estimation also reveals the relationship between the approximation rate of FNNs and the topology structure of hidden layer.
引用
收藏
页码:1321 / 1327
页数:7
相关论文
共 32 条
[1]  
[Anonymous], 1994, SCI CHINA A
[2]   UNIVERSAL APPROXIMATION BOUNDS FOR SUPERPOSITIONS OF A SIGMOIDAL FUNCTION [J].
BARRON, AR .
IEEE TRANSACTIONS ON INFORMATION THEORY, 1993, 39 (03) :930-945
[3]   The estimate for approximation error of neural networks: A constructive approach [J].
Cao, Feilong ;
Xie, Tingfan ;
Xu, Zongben .
NEUROCOMPUTING, 2008, 71 (4-6) :626-630
[4]  
CAO FL, 2008, ACTA MATH SINICA, V51, P91
[5]  
CAO FL, 2007, ACTA MATH SINICA, V50, P385
[6]   UNIVERSAL APPROXIMATION TO NONLINEAR OPERATORS BY NEURAL NETWORKS WITH ARBITRARY ACTIVATION FUNCTIONS AND ITS APPLICATION TO DYNAMICAL-SYSTEMS [J].
CHEN, TP ;
CHEN, H .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 1995, 6 (04) :911-917
[7]   APPROXIMATION CAPABILITY TO FUNCTIONS OF SEVERAL VARIABLES, NONLINEAR FUNCTIONALS, AND OPERATORS BY RADIAL BASIS FUNCTION NEURAL NETWORKS [J].
CHEN, TP ;
CHEN, H .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 1995, 6 (04) :904-910
[8]   Improved rates and asymptotic normality for nonparametric neural network estimators [J].
Chen, XH ;
White, H .
IEEE TRANSACTIONS ON INFORMATION THEORY, 1999, 45 (02) :682-691
[9]   APPROXIMATION BY RIDGE FUNCTIONS AND NEURAL NETWORKS WITH ONE HIDDEN LAYER [J].
CHUI, CK ;
LI, X .
JOURNAL OF APPROXIMATION THEORY, 1992, 70 (02) :131-141
[10]  
Cybenko G., 1989, Mathematics of Control, Signals, and Systems, V2, P303, DOI 10.1007/BF02551274