Classification Methods for Remotely Sensed Data, Second Edition

Page 106

Figure 3.2 Example of forward propagation procedure on neurone j.

of search methods can be applied to find the minimum of the function E(w). One widely used method is that of steepest descent, in which each weight is updated in terms of the expression:

where is the weight associated with the connection from neurone i to j at time n, η (the learning rate) is a small positive constant, in the range 0≤η≤1, controlling the size of the step, and is the gradient at point , which can be interpreted as the change in error E that results from a change in weight . The initial values of the weights, , are