Pseudoinverse techniques, information theory, and the training of feedforward networks
Luis Anibal Diambra, J. Fernández, Angel Ricardo Plastino · Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics · 1995
An information theory based method for the training of perceptrons is presented. Our technique guarantees an errorless learning process for learnable mappings with just a minimum amount of examples. The only requirement is that the transfer function must possess an inverse. Some illustrative results are presented. The method can be considered to yield another tool for feed-forward training.