R70-8 Analog Methods for On-Line System Identification Using Noisy Measurements

A stochastic approximation, or stochastic gradient procedure, is a method of estimating the minimum of a criterion function despite the random effects of noise. It was originally conceived by the mathematician as a means of estimating the minimum of a regression function. In recent years, however, stochastic approximation has received considerable attention in the engineering field because of its potential application to learning systems, parameter estimation, and automatic control.