Promotion-based input partitioning of neural network

To improve the learning performance and precision of neural network, this paper introduces an input-attribute partitioning algorithm with an aim to increase the promotion among them. If a better performance could be obtained by training some attributes together, it is considered that there is positive effect among these attributes. It is assumed that by putting attributes, among which there are positive effect, a lower error can be obtained. After partitioning, multiple learners were employed to tackle each group. The final result is obtained by integrating the result of each learner. It turns out that, this algorithm actually can reduce the classification error in supervised learning of neural network. ? Springer-Verlag Berlin Heidelberg 2014.