ANALOG VLSI AND MULTILAYER PERCEPTRONS - ACCURACY, NOISE AND ON-CHIP LEARNING

被引：2

作者：

MURRAY, AF ^{[1
]}

机构：

[1] UNIV EDINBURGH,DEPT ELECT ENGN,EDINBURGH EH8 9YL,MIDLOTHIAN,SCOTLAND

来源：

NEUROCOMPUTING | 1992年 / 4卷 / 06期

关键词：

ANALOG VLSI; BACKPROPAGATION LEARNING; MULTILAYER PERCEPTRONS; NOISE IMMUNITY; VIRTUAL TARGETS LEARNING SCHEME;

D O I：

10.1016/0925-2312(92)90015-H

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

The work described in this paper aims to distil the useful features of backpropagation, along with a measure of analog pragmatism, to produce a learning scheme suitable for on-chip VLSI implementation. As the so-called 'virtual target' algorithm has its roots in the same concepts as backpropagation, it shares features, and has similar performance. Some experimental results are shown and a preliminary analog learning chip architecture is proposed.

引用

页码：301 / 310

页数：10

共 7 条

[1] GROSSMANN T, 1989, NEURAL INFORMATION P, P73
[2] PULSE ARITHMETIC IN VLSI NEURAL NETWORKS
MURRAY, AF
[J]. IEEE MICRO, 1989, 9 (06) : 64 - 74
[3] PULSE-STREAM VLSI NEURAL NETWORKS MIXING ANALOG AND DIGITAL-TECHNIQUES
MURRAY, AF
DELCORSO, D
TARASSENKO, L
[J]. IEEE TRANSACTIONS ON NEURAL NETWORKS, 1991, 2 (02): : 193 - 204
[4] MURRAY AF, 1989, NEURAL INFORMATION P
[5] ROHWER R, 1989, NEURAL INFORMATION P, P558
[6] Rumelhart David E., 1987, LEARNING INTERNAL RE, P318
[7] WATTS S, THESIS U OXFORD

← 1 →