Reduced pattern training in pattern distributor networks

In this paper, we propose a new task decomposition method, Task Decomposition with Pattern Distributor (PD), for multilayered feedforward neural networks. The method uses a combination of network modules in parallel and series to generate the overall solution for a complex problem. We also introduce a method called reduced pattern training in PD networks. This method aims to improve the performance of the pattern distributor network. Our analysis and the experimental results show that reduced pattern training improves the performance of pattern distributor network significantly. Experimental results confirm that this new method can reduce training time and improve network generalization accuracy significantly when compared to ordinary task decomposition methods such as Output Parallelism.