Page 41 - Artificial Intelligence in the Age of Neural Networks and Brain Computing
P. 41

28     CHAPTER 1 Nature’s Learning Rule: The Hebbian-LMS Algorithm























                         FIGURE 1.18
                         A network with three Hebbian-LMS hidden layers and a trainable LMS output layer with a
                         one-out-of-many code.



























                         FIGURE 1.19
                         Learning curves obtained in statistical classification experiments with a Hebbian-LMS
                         neural network similar to the one illustrated in Fig. 1.18.


                         trained with Hebbian-LMS. The output layer was trained with LMS to implement a
                         one-out-of-many code. The learning curves are plots of output error versus the
                         number of training cycles, each cycle being a training sequence including the entire
                         set of training patterns. The output error is a count of number of errors or misclas-
                         sifications. Plots were given for both training and testing datasets.
   36   37   38   39   40   41   42   43   44   45   46