Analog versus discrete neural networks

被引：8

作者：

DasGupta, B ^{[1
]}

Schnitger, G ^{[1
]}

机构：

[1] UNIV FRANKFURT,FACHBEREICH INFORMAT 20,D-60054 FRANKFURT,GERMANY

来源：

NEURAL COMPUTATION | 1996年 / 8卷 / 04期

关键词：

D O I：

10.1162/neco.1996.8.4.805

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

We show that neural networks with three-times continuously differentiable activation functions are capable of computing a certain family of n-bit Boolean functions with two gates, whereas networks composed of binary threshold functions require at least Omega(log n) gates. Thus, for a large class of activation functions, analog neural networks can be more powerful than discrete neural networks, even when computing Boolean functions.

引用

页码：805 / 818

页数：14

共 14 条

[1]

BOPPANA R, 1990, HDB THEORETICAL COMP

[2]

DASGUPTA B, 1993, ADV NEURAL INFORMATI, V5, P615

[3]

GOLDMANN M, 1991, COMPUT COMPLEX, V1, P113

[4]

Hajnal A., 1987, 28th Annual Symposium on Foundations of Computer Science (Cat. No.87CH2471-1), P99, DOI 10.1109/SFCS.1987.59

[5] COMPUTATIONAL LIMITATIONS ON TRAINING SIGMOID NEURAL NETWORKS [J].

HOFFGEN, KU .

INFORMATION PROCESSING LETTERS, 1993, 46 (06) :269-274

[6]

Maass W., 1993, Proceedings of the Twenty-Fifth Annual ACM Symposium on the Theory of Computing, P335, DOI 10.1145/167088.167193

[7]

MAASS W, 1991, 32ND P IEEE S F COMP, P767

[8]

Macintyre A., 1993, Proceedings of the Twenty-Fifth Annual ACM Symposium on the Theory of Computing, P325, DOI 10.1145/167088.167192

[9] LOWER BOUNDS ON THE SIZE OF BOUNDED DEPTH CIRCUITS OVER A COMPLETE BASIS WITH LOGICAL ADDITION [J].

RAZBOROV, AA .

MATHEMATICAL NOTES, 1987, 41 (3-4) :333-338

[10]

REIF J, 1987, 2ND P STRUCT COMPL T, P118

← 1 2 →