Catalogue of Artificial Intelligence Techniques
Keywords: Hebbian learning, learning, neural networks
Categories: Neural Networks
Author(s): Ashley Walker
The Willshaw network was designed to model the regular structure of nerve nets in the central nervous system. It is composed of a matrix of binary synapses (i.e., weighted connections) that feed into binary output units--whose function it is to threshold the sum of the input signals. The Willshaw network is used as an Associative Memory, i.e., it maps inputs onto outputs, and, as such, operates more efficiently than the Hopfield network. The Willshaw net employs a training rule which is a variant of Hebbian learning. . For each pair of association patterns, the appropriate bit patterns are simultaneously presented along input and output lines and synapses are turned on where input and output lines are coactive. (Synapses which are not used in pattern learning stay off.) During recall, the input pattern is presented as before, and each output unit sums the contribution from all activated synapses connecting it to active lines in the input pattern. In a fully connected Willshaw net, this sum is then thresholded according to the number of bits turned on in the original pattern, however, for partially connected nets, the thresholding strategy becomes more complicated.
No references to display.