KOLMOGOROV THEOREM AND MULTILAYER NEURAL NETWORKS

被引：473
作者：
KURKOVA, V
机构：
来源：
NEURAL NETWORKS | 1992年 / 5卷 / 03期
关键词：
FEEDFORWARD NEURAL NETWORKS; MULTILAYER PERCEPTRON TYPE NETWORKS; SIGMOIDAL ACTIVATION FUNCTION; APPROXIMATIONS OF CONTINUOUS FUNCTIONS; UNIFORM APPROXIMATION; UNIVERSAL APPROXIMATION CAPABILITIES; ESTIMATES OF NUMBER OF HIDDEN UNITS; MODULUS OF CONTINUITY;
D O I：
10.1016/0893-6080(92)90012-8
中图分类号：
TP18 [人工智能理论];
学科分类号：
081104 ; 0812 ; 0835 ; 1405 ;
摘要：
Taking advantage of techniques developed by Kolmogorov, we give a direct proof of the universal approximation capabilities of perceptron type networks with two hidden layers. From our proof we derive estimates of numbers of hidden units based on properties of the function being approximated and the accuracy of its approximation.;
引用
收藏
页码：501 / 506
页数：6
相关论文
共 14 条
[1]  
ARNOLD VI, 1957, DOKL AKAD NAUK SSSR+, V114, P679
[2]  
Carroll S M, 1989, P IJCNN P, P607
[3]  
Cybenko G., 1989, Mathematics of Control, Signals, and Systems, V2, P303, DOI 10.1007/BF02551274
[4]   ON THE APPROXIMATE REALIZATION OF CONTINUOUS-MAPPINGS BY NEURAL NETWORKS [J].
FUNAHASHI, K .
NEURAL NETWORKS, 1989, 2 (03) :183-192
[5]   Representation Properties of Networks: Kolmogorov's Theorem Is Irrelevant [J].
Girosi, Federico ;
Poggio, Tomaso .
NEURAL COMPUTATION, 1989, 1 (04) :465-469
[6]  
Hecht-Nielsen R., 1989, IJCNN: International Joint Conference on Neural Networks (Cat. No.89CH2765-6), P593, DOI 10.1109/IJCNN.1989.118638
[7]  
Hecht-Nielsen R., 1990, NEUROCOMPUTING
[8]  
HECHTNIELSEN R, 1987, 1ST IEEE INT C NEUR, V3, P11
[9]   APPROXIMATION CAPABILITIES OF MULTILAYER FEEDFORWARD NETWORKS [J].
HORNIK, K .
NEURAL NETWORKS, 1991, 4 (02) :251-257
[10]   MULTILAYER FEEDFORWARD NETWORKS ARE UNIVERSAL APPROXIMATORS [J].
HORNIK, K ;
STINCHCOMBE, M ;
WHITE, H .
NEURAL NETWORKS, 1989, 2 (05) :359-366
12