Abstract
Recently several articles have discussed the advantages of higher-order neural networks.1 These include shorter learning times, increased memory storage capacities, and more flexible pattern decision boundaries. Problems that cannot be solved by linear neural networks without hidden layers can be easily solved by looking at higher- order correlations in the input data vector. A classic example is the EX-OR (exclusive OR) problem. Our previous research in optical polynomial processing2 has led us to investigate an optical implementation for the quadratic neural network. This research has recently resulted in a highly parallel architecture which is well suited to time multiplexing and/or space multiplexing to achieve a large neural network.
© 1988 Optical Society of America
PDF ArticleMore Like This
Alex V. Huynh, John F. Walkup, and Thomas F. Krile
ThY35 OSA Annual Meeting (FIO) 1990
T. L. Jong, Stephen G. Batsell, John F. Walkup, and Thomas F. Krile
THHH3 OSA Annual Meeting (FIO) 1988
Alex V. Huynh, John F. Walkup, and Thomas F. Krile
MII8 OSA Annual Meeting (FIO) 1991