Page 449 - Introduction to Information Optics

434 7. Pattern Recognition with Optics

0 1
1 1

B 1
OUTPUT

Fig. 7.68.

(d) Show that the partial input of an Arabic numeral will produce the
correct Roman numeric translation.
7.45 (a) What are the major differences in operation between the Kohonen
and the Hopfield neural nets?
(b) Describe the effects and the strategies for selecting the neighborhood
N c(t) in a Kohonen neural net.
7.46 Assume that the initial memory matrix in a Kohonen neural net is
distributed by the 50% random noise shown in Fig. 7.53(b). A training
set of Roman letters {A, B, C, X] is to be stored in the self-organizing
neural net, in which a linear learning speed is used, where a(0) = 0.02 and
a = 0.00005.
(a) Simulate the memory matrix space after 200 iterations.
(b) Repeat part (a) for 400 iterations.
(c) Compare the results of parts (a) and (b).
7.47 Refer to the polychromatic neural network of Fig., 7.58.
(a) Simulate a polychromatic heteroassociation IWM for pattern trans-
lations.
(b) Show (by simulation) if a partial input of "red" A will result in the
correct translation of a "red" Chinese character, as shown in this
figure.