Piecewise-linear neural network - possible training algorithms efficiency comparison

Abstract

In this article, a benchmark of algorithms for training of piecewise-linear artificial neural networks is introduced. At first, motivation of this article is described for a special topology of the neural network is used. This topology can be advantageously used in system control design, but it is difficult problem to train it. In this article, there is described a set of possible training algorithms, these algorithms are tested and evaluated. Benchmarking data are based on real problems.