Abstract
A novel connectionist architecture that differs from conventional architectures based on the neuroanatomy of biological organisms is described. The proposed scheme is based on the model of multilayered optical thin-films, with the thicknesses of the individual thin-film layers serving as adjustable ‘weights’ for the training. A discussion of training techniques for this model and some sample simulation calculations in the area of pattern recognition are presented. These results are shown to compare with results when the same training data are used in connection with a feed-forward neural network with back propagation training. A physical realization of this architecture could largely take advantage of existing optical thin-film deposition technology.