Abstract

Neural networks have become attractive resources for the solution of a large class of problems which are not easily handled by traditional computing tools and strategies[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]. The massive sharing of information between nodal processors in neural network configurations allows both redundancy for fault tolerance and a high level of association of data for classification problems such as sorting and character or pattern recognition. It is this massive data distribution that is at once the advantage of neural computing in that it facilitates solutions to difficult information processing problems and a disadvantage in that the enormous number of required interconnection paths can severely limit the size or capability of the network if wire paths are used.

© 1989 Optical Society of America