APPLICATION OF THE ANNA NEURAL NETWORK CHIP TO HIGH-SPEED CHARACTER-RECOGNITION

被引：45

作者：

SACKINGER, E

BOSER, BE

BROMLEY, J

LECUN, Y

JACKEL, LD

机构：

[1] AT&T Bell Laboratories, Holmdel

来源：

IEEE TRANSACTIONS ON NEURAL NETWORKS | 1992年 / 3卷 / 03期

关键词：

D O I：

10.1109/72.129422

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

A neural network with 136 000 connections for recognition of handwritten digits has been implemented using a mixed analog/digital neural network ship. The neural network chip is capable of processing 1000 characters per second. The recognition system has essentially the same error rate (5%) as a simulation of the network with 32-bit floating-point precision.

引用

页码：498 / 505

页数：8

共 8 条

[1]

BOSER B, 1991, ISSCC, P184

[2] AN ANALOG NEURAL NETWORK PROCESSOR WITH PROGRAMMABLE TOPOLOGY [J].

BOSER, BE ;

SACKINGER, E ;

BROMLEY, J ;

LECUN, Y ;

JACKEL, LD .

IEEE JOURNAL OF SOLID-STATE CIRCUITS, 1991, 26 (12) :2017-2025

[3]

GRAF HF, 1991, P INT JOINT C NEURAL, P481

[4]

GUYON I, 1990, PATTERN RECOGNITION

[5]

KANTER I, 1990, COMMUNICATION

[6]

LECUN Y, 1990, NEURAL INFORMATION P, P396

[7]

Vapnik V, 1982, ESTIMATION DEPENDENC

[8] PHONEME RECOGNITION USING TIME-DELAY NEURAL NETWORKS [J].

WAIBEL, A ;

HANAZAWA, T ;

HINTON, G ;

SHIKANO, K ;

LANG, KJ .

IEEE TRANSACTIONS ON ACOUSTICS SPEECH AND SIGNAL PROCESSING, 1989, 37 (03) :328-339

← 1 →