Analysis of Learning Processes of Chaotic Time Series by Neural Networks
Tsuyoshi Hondou, Yasuji Sawada · Progress of Theoretical Physics · 1994
Study of the learning process of chaotic time series by neural networks with back-propagation algorithm showed the existence of a general learning process. The process was found to be efficiently characterized by newly defined normalized measures, the coherence and the direction-cosine for the motion of weight vectors.