Fault immunization model and algorithm for a self-organizing artificial neural network

Abstract

Self-Organizing Mapping (SOM) neural network has been widely used in pattern classification, vector quantization, and image compression. We consider the problem of strengthening the reliability of an SOM neural network by applying a fault immunization technique to each neuron synaptic links, which is similar to the concept of biological immunization. Instead of assuming the stuck-at-0 and stuck-at-1 as in the previous studies, we consider a general case of stuck-at-a, where a is a real value. Our only assumption is that only one neuron can be faulty at any time. There is no restriction on the number of faulty links of the neuron. Let Wi,j be the weight of synaptic link j of neuron i obtained after the winner-take-all classification. Weight Wi,j is immunized by adding a constant sigmai,j, either positive or negative, to Wi,j. A neuron reaches its maximum fault immunization if the value of Wi,j+sigmai,j can be either increased or decreased as much as possible without creating any misclassification. Thus, the fault immunization problem is formulated as an optimization problem on findings the value of each sigmai,j. A technique to find the value of Wi,j+sigmai,j and its application to enhance the transmission reliability in image compression area is proposed in this thesis.