Current supervised learning techniques require a sequential feeding of the training input vectors and their targets. This process is very time consuming and convergence is difficult to control due to the ignorance of the nature of the training vectors. We therefore propose another approach to solving the training problem by analyzing the training vectors and cluster them into groups. Each group then can be learned in parallel. In this paper, we focus our study only on one dimensional, real space input vectors and the class of single-input single-output feedforward neural network. However, the parallel concept developed in this paper can possibly be extended to a higher dimensional space.
展开▼
