A Matlab Tool for Analyzing and Improving Fault Tolerance of Artificial Neural Networks

AbstractFTSET is a software tool that deals with fault tolerance of Artificial Neural Networks. This tool is capable of evaluating the fault tolerance degree of a previously trained Artificial Neural Network given its inputs ranges, the weights and the architecture. The FTSET is also capable of improving the fault tolerance by applying a technique of splitting the connections of the network that are more important to form the output. This technique improves fault tolerance without changing the network's output. The paper is concluded by two examples that show the application of the FTSET to different Artificial Neural Networks and the improvement of the fault tolerance obtained.