Convex error functions for multilayered perceptrons

Summary form only given. For multilayered perceptrons, the error is generally defined as a quadratic function of the difference between the ideal and real outputs. Unfortunately, this L² error, considered as a function of the weights, is generally not convex. The author has studied the networks without a hidden layer and has constructed a convex error function which is also a Kullback information between two `natural´ probabilities on the outputs. It is shown how this error function may be practically used for more general perceptrons