Evolution of fuzzy uncertainty in neural network learning
Salahalddin T. Abusalah · 2002
The paper explores the development of artificial neural network learning dynamics in terms of fuzzy uncertainty. In conventional artificial neural networks utilizing crisp variables, a set of error metrics required to achieve network convergence can be developed in the information-theoretic plane (based on the probabilistic uncertainty of the network variables). However, in a fuzzy neural network, consideration of fuzzy uncertainties can also facilitate a model to depict the convergence dynamics in the information-theoretic plane. A formulation is presented to achieve the fusion of a fuzzy neural network with the information-theoretic cost functions.