Backpropagation Applied to Handwritten Zip Code Recognition

被引:8142
作者
LeCun, Y. [1 ]
Boser, B. [1 ]
Denker, J. S. [1 ]
Henderson, D. [1 ]
Howard, R. E. [1 ]
Hubbard, W. [1 ]
Jackel, L. D. [1 ]
机构
[1] AT&T Bell Labs, Holmdel, NJ 07733 USA
关键词
D O I
10.1162/neco.1989.1.4.541
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The ability of learning networks to generalize can be greatly enhanced by providing constraints from the task domain. This paper demonstrates how such constraints can be integrated into a backpropagation network through the architecture of the network. This approach has been successfully applied to the recognition of handwritten zip code digits provided by the U.S. Postal Service. A single network learns the entire recognition operation, going from the normalized image of the character to the final classification.
引用
收藏
页码:541 / 551
页数:11
相关论文
共 15 条
[1]   What Size Net Gives Valid Generalization? [J].
Baum, Eric B. ;
Haussler, David .
NEURAL COMPUTATION, 1989, 1 (01) :151-160
[2]  
BECKER S, 1988, CRGTR885 U TOR
[3]  
Bottou L.-Y., 1988, P NEURONIMES 88 NIM
[4]  
Denker J., 1987, Complex Systems, V1, P877
[5]  
Denker J. S., 1989, ADV NEURAL INFORM PR
[7]   RECEPTIVE FIELDS, BINOCULAR INTERACTION AND FUNCTIONAL ARCHITECTURE IN CATS VISUAL CORTEX [J].
HUBEL, DH ;
WIESEL, TN .
JOURNAL OF PHYSIOLOGY-LONDON, 1962, 160 (01) :106-&
[8]  
Lecun Y., 1987, THESIS U P M CURIE P, V6
[9]  
LeCun Y., 1989, CONNECTIONISM PERSPE
[10]  
MOZER MC, 1987, ATTENTION PERFORM, P83