This report describes a neural network architecture ClusNet designed for the prediction of chaotic time series. It advantages include simplicity, fast and sure convergence, and less need for computing resources. After describing its architecture and learning algorithms, its prediction perfo1:mance on the Logistic and the Mackay-Class chaotic time series is presented. Compared to other current prediction approaches, ClusNet predicts with the same level of accuracy while utilizing less resources.

Date of this Version

September 1992