On the error function of Interval Arithmetic Backpropagation

摘要

A generalization of Multilayer Feedforward and Backpropagation to interval arithmetic was proposed several years ago. This generalization has several applications like the codification of expert's knowledge in the form of rules, "don't care attributes", missing inputs, etc. In the bibliography there are two error function and two training algorithms proposed, and in this paper we present the first comparison between both training algorithms. We have used a total number of 35 different problems for the comparison, Finally, the results are that one of the two training algorithms clearly out-perform the other.