Maximum entropy, pseudoinverse techniques, and time series predictions with layered networks
Luis Anibal Diambra, Angelo Plastino · Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics · 1995
A maximum-entropy-based method for the training of layered networks is presented. Our technique guarantees an errorless learning process for learnable mappings with just a minimum number of examples. The network is proposed for nonlinear systems prediction. Some numerical examples for chaotic time series are presented. The method can be considered to yield an alternative tool for feed-forward training.