Hierarchical neural networks for time-series analysis and control
T Fröhlinghaus, Andreas Weichert, Pál Ruján · Network Computation in Neural Systems · 1994
A hierarchical network of neurons with local receptive fields is constructed by using a mixture of clustering and supervised learning strategies. The network's performance is enhanced by built-in importance sampling and by its multiscale organization in both time and phase space domains. The method is tested on predicting chaotic time-series and on balancing a physical pendulum.