A ClusNet architecture for prediction

A neural network architecture, ClusNet, for accurately predicting the temporal continuation of a time series is described. This network is designed to retain much of the advantages associated with instance-based methods. Prediction results obtained with ClusNet are compared with other popular neural network methods. Without the need for tuning, ClusNet achieves reasonable results using a fraction of the resources required by instance-based methods in the Mackey-Glass time series prediction problem. ClusNet is ideal as a rapid prototyping tool for applications in which fast online learning is required. Its convergence can be proven and in the authors´ experiments, it converges quickly