Page 428 - Introduction to Information Optics
P. 428

7.8. Neural Pattern Recognition




















            Fig. 7.47. Heteroassociation. S, a and S 2 are the input and output pattern spaces.



       subjected to inhibition from both input neurons 3 and 4, and so on. The
       corresponding heteroassociation memory matrix is shown in Fig. 7.48c, for
       which 1,0. — 1 represents the excitory, null, and inhibitory connections. The
       memory matrix is partitioned into four blocks, and each block represents the
       corresponding output neurons. For illustration, we show a supervised hetero-
       association NN, as obtained from the optical NN of Fig. 7.7. A set of Arabic



                                       a i
                                        1  0

                                        (a)

                  nput Neurons  Output Neurons










                      Interconnections
                      —— Excitation
                      — — inhibition
                                                         (c)
                           (b)
       Fig. 7.48. Construction of an heteroasociative IWM (HA IWM). (a) Input-output training sets,
       (b) A tristate NN, (c) Heteroassociation IWM.
   423   424   425   426   427   428   429   430   431   432   433