A learning rule in the Chebyshev norm for multilayer perceptrons

An <e1>L</e1><sub>∞</sub> version of the backpropagation paradigm is proposed, and a comparison between the <e1>L</e1><sub>2 </sub>and the <e1>L</e1><sub>∞</sub> paradigms is presented, taking into account computational cost and speed of convergence. The convergence of the <e1>L</e1><sub>∞</sub> algorithm has been successfully tested in several cases