The competitive selection of artificial neural network training sets using an arms race model

The selection of artificial neural network training sets can be problematical in some situations. This paper presents a novel method of developing such datasets. Ideas from the arms race between competing superpowers are used to develop a robust technique for an artificial neural network training set selection. Two modules are used, one to select candidates for the training set, the second to train an ANN on the selected dataset. The results of the learning process are used to modify the training set selection accordingly. An example to train an ANN for electrocardiogram (ECG) classification is presented to demonstrate the concept.