The effects of segmentation on back-propagation networks

Segmented neural networks and their capabilities when used to process several types of data are examined. The effect that segmentation has upon the network´s topology and the learning rule is investigated. The training method used is a variation of the backpropagation (BP) rule. Network segmentation causes some variation in the behavior of the learning rule. Modifications to the BP rule are also examined which illustrate how it can be improved for use with a segmented topology. Testing involves comparisons of segmented and unsegmented networks in an attempt to identify the effects of delays caused by the segmentation of the neural components. Comparisons are made between the rate of learning and recall, accuracy, and capacity for several configurations of a BP network