Interval GA for evolving neural networks with interval weights and biases.

In this paper, we propose an extension of genetic algorithm for neuroevolution of interval-valued neural networks. In the proposed GA, values in the genotypes are not real numbers but intervals. We apply our interval-valued GA (IvGA) to the approximate modeling of interval functions with interval-valued neural networks. Experimental results showed that INNs trained by our IvGA approximated a test function to a certain extent, despite the fact that the learning was not supervised. [ABSTRACT FROM PUBLISHER]