Interval-valued evolution strategy for evolving neural networks with interval weights and biases
Hidehiko Okada, Tetsuya Wada, Akira Yamashita, Takashi Matsuse · 2012
In this paper, we propose an extension of evolution strategy (ES) for evolving interval-valued neural networks. In the proposed ES, values in the genotypes are not real numbers but intervals. We apply our interval-valued ES (IES) to the approximate modeling of interval functions with interval-valued neural networks (INNs). Experimental results showed that INNs trained by our IES could well approximate a hidden test function, despite the fact that the learning was not supervised.