DocumentCode :
1431214
Title :
Fault tolerance via weight noise in analog VLSI implementations of MLPs-a case study with EPSILON
Author :
Edwards, Peter J. ; Murray, Alan F.
Author_Institution :
Dept. of Electr. Eng., Edinburgh Univ., UK
Volume :
45
Issue :
9
fYear :
1998
fDate :
9/1/1998 12:00:00 AM
Firstpage :
1255
Lastpage :
1262
Abstract :
Training with weight noise has been shown to be an effective means of improving the fault tolerance of multilayer perceptrons (MLPs). This paper investigates the use of weight noise used during MLP training to compensate for the inherent errors encountered in VLSI implementations. Weight adaptation is conducted solely in software, eliminating the need for costly in-the-loop training. The particular VLSI implementation considered here is the EPSILON processor card developed at Edinburgh University. Both software and hardware experiments demonstrate the effectiveness of this approach. This case study with the EPSILON processor card highlights what we believe to be a number of common inadequacies with custom designed hardware. In particular, the limitations of EPSILON in terms of its dynamic range performance has been shown to be a problem. In summary, we show that networks trained with weight noise are fault-tolerant but also require an increased dynamic range to exploit this property
Keywords :
VLSI; analogue integrated circuits; analogue processing circuits; fault tolerant computing; integrated circuit noise; learning (artificial intelligence); multilayer perceptrons; neural chips; EPSILON processor card; MLP training; analog VLSI; dynamic range; fault tolerance; multilayer perceptron; weight noise; Application software; Circuit faults; Computer aided software engineering; Dynamic range; Fault tolerance; Fault tolerant systems; Multilayer perceptrons; Neural network hardware; Neural networks; Very large scale integration;
fLanguage :
English
Journal_Title :
Circuits and Systems II: Analog and Digital Signal Processing, IEEE Transactions on
Publisher :
ieee
ISSN :
1057-7130
Type :
jour
DOI :
10.1109/82.718593
Filename :
718593
Link To Document :
بازگشت