On the convergence behavior of Rosenblatt's perceptron learning algorithm

摘要

A stochastic analysis of the steady-state and transient convergence properties of a single-layer perceptron is presented. The training data are modeled using a system identification formulation with Gaussian inputs, and the perceptron weights are adjusted by Rosenblatt's learning algorithm. It is shown that the convergence points of the algorithm depend on the step size mu and the input signal power sigma /sub x//sup 2/. Two coupled nonlinear recursions that describe the transient behavior of the algorithm are derived. Computer simulations that verify the analytical models are also presented.