Neural Network Computing for the Electric Power Industry

B. Neural Network Methodologies

Backpropagation and its Applications

Bernard Widrow

Michael A. Lehr

Stanford University Department of Electrical Engineering, Stanford, CA 94305-4055

Backpropagation remains the most widely used neural network algorithm. We present a basic review of this method and its applications. The backpropagation algorithm’s roots are also explored in a discussion of its predecessor, the LMS Algorithm.

Introduction

The field of neural networks has enjoyed major advances since 1960, a year which saw the introduction of two of the earliest feedforward neural network algorithms: the Perceptron rule [1] and the LMS algorithm (or Widrow-Hoff rule) [2]. Around 1961, Widrow and his students devised Madaline Rule I (MRI), the ...