Page 162 - Introduction to Information Optics

References 14:

Training Set

Input Hetero-association Output
Neural Network

Fig. 2.57. A block box representation of a heteroassociation NN.

about 4 stored letters, whereas the I PA model is quite stable for all 26 letters.
Even for a 10% input noise level, it can retrieve effectively up to 12 stored
letters. As for noiseless input, the IPA model can, in fact, produce correct
results for all 26 stored letters.
Furthermore, pattern translation can also be accomplished using a hetero-
association IPA neural network. Notice that by using similar logic operations
among input-output (translation) patterns, a heteroassociative IWM can be
constructed. An example of the heteroassociation NN is shown in Fig. 2.57,
The objective is to translate a set of English alphabets to a set of Chinese
characters. We see that if A is presented to the heteroassociation NN, a
corresponding Chinese character is converged at the output end.

REFERENCES

i F. T. S. Yu and S. Jutamulia, Optical Signal Processing, Computing and Neural Networks.
Wiley-Interscience, New York, (1992).
2. A. VanderLugt, Signal Detection by Complex Spatial Filtering, IEEE Trans. Inform. Theory,
IT-10, 139(1965).
3. L. j. Cutrona et al., On the Application of Coherent Optical Processing Techniques to Synthetic
Aperture Radar, Proc. IEEE, 54, 1026 (1966).
4. D. Casacent and D. Psaltis, Scale Invariant Optical Correlation Using Mullin Transforms, Opt.
Commun., 17, 59 (1976).
5. C. F. Hester and D. Casacent, Multivariant Technique for Multiclass Pattern Recognition, Appl.
Opt., 19, 1758 (1980).