Detecting constructions of nonlinear integral systems from input-output data: an application of neural networks

Abstract

If the input-output relation of a multi-input system can be represented by some kind of integral with respect to a nonnegative monotone set function, which is not necessarily additive, then the construction of the system may be entirely described by the monotone set function. After obtaining input-output data from such a system, the set function can be optimally determined by using a specially designed neural network algorithm.