On the Evolutionary Neural Network Creation Using Native Visibility Graph

In this article, a novel way of creating the feed forward neural network is presented. This model is inspired by native visibility graph, and the networks are created by well-known differential evolution algorithm. In the experiment parts, we compared these networks with multi-layer perceptron. We have found that our novel approach creates smaller networks comparing to multi-layer perceptron and also, the decision space differs. The decision space was similar for multi-layer perceptron for almost all training runs from 51 repetitions, but for our novel approach, the decision space differs and also creates more complex patterns.