Threshold effects in an error-correction learning algorithm

The threshold-function effects on the changes in a feedback gain vector, made during an errorcorrection updating procedure, have been demonstrated. The qualities referred to earlier are given as a transition diagram figure. With a threshold function present, the feedback gains may change in any one or more of six different ways. On the other hand, in the absence of a threshold function, the feedback gains always change in the same manner, either by increasing, decreasing, increasing, etc., or by decreasing, increasing, decreasing, etc. The limit cycles in the figure are a general property of error-correction learning algorithms in which the feedback gains are subject to gradient-type changes.