A transient learning comparison of Rosenblatt, backpropagation, and LMS algorithms for a single-layer perceptron for system identification

Author

Engel, Isaac ; Bershad, Neil J.

Author_Institution

Dept. of Electr. & Comput. Eng., California Univ., Irvine, CA, USA

Volume

42

Issue

5

fYear

1994

fDate

5/1/1994 12:00:00 AM

Firstpage

1247

Lastpage

1251

Abstract

Presents a transient performance comparison of the Rosenblatt (1962), backpropagation, and LMS training algorithms for a single layer perceptron. The perceptron is attempting to identify a specific nonlinear system with Gaussian inputs. A summary of the first and second moment behaviors of the three algorithms is presented. With the criteria of probability of correct classification versus the number of algorithm iterations, the statistical results are used to compare the learning performance of the algorithms for some specific parameter values. By both the theoretical analysis and by Monte Carlo simulations, it is shown that there are no significant learning performance differences among these three algorithms for these parameter selections

Keywords

backpropagation; identification; least squares approximations; linear systems; neural nets; probability; stochastic processes; Gaussian inputs; LMS algorithm; Monte Carlo simulations; Rosenblatt algorithm; algorithm iterations; backpropagation algorithm; correct classification probability; first moment behavior; learning performance; nonlinear system; second moment behavior; single-layer perceptron; statistical results; system identification; transient learning; transient performance comparison; Adaptive algorithm; Algorithm design and analysis; Backpropagation algorithms; Convergence; Least squares approximation; Nonlinear systems; Performance analysis; Probability; Signal processing algorithms; System identification;

fLanguage

English

Journal_Title

Signal Processing, IEEE Transactions on

Publisher

ieee

ISSN

1053-587X

Type

jour

DOI

10.1109/78.295190

Filename

295190