Learning of Associative Memory Networks Based Upon Cone-Like Domains of Attraction

A learning algorithm for single layer perceptrons is proposed. First, cone-like domains, each of which i s mapped by the perceptron network into almost an associative pattern, are derived. The learning algorithm is obtained as a process that enlarges the cone-like domains. For autoassociative networks, it is shown that the cone-like domains become domains of attraction for stored patterns in the network. In this case, extended domains of attraction are also obtained by feeding t h e outputs of the network back to the input layer. I n computer simulations, character recognition ability of the autoassociative network i s examined