A Computational Framework for Efficient Error Correcting Codes Using an Artificial Neural Network Paradigm.

The quest for an efficient computational approach to neural connectivity problems has undergone a significant evolution in the last few years. The current best systems are far from equaling human performance, especially when a program of instructions is executed sequentially as in a von Neuman computer. On the other hand, neural net models are potential candidates for parallel processing since they explore many competing hypotheses simultaneously using massively parallel nets composed of many computational elements connected by links with variable weights. Thus, the application of modeling of a neural network must be complemented by deep insight into how to embed algorithms for an error correcting paradigm in order to gain the advantage of parallel computation. In this dissertation, we construct a neural network for single error detection and correction in linear codes. Then we present an error-detecting paradigm in the framework of neural networks. We consider the problem of error detection of systematic unidirectional codes which is assumed to have double or triple errors. The generalization of network construction for the error-detecting codes is discussed with a heuristic algorithm. We also describe models of the code construction, detection and correction of t-EC/d-ED/AUED (t-Error Correcting/d-Error Detecting/All Unidirectional Error Detecting) codes which are more general codes in the error correcting paradigm