On Hypercomplex Networks

TL;DR

This paper introduces hypercomplex networks (HC) characterized by heterogeneity across multiple topological measurements, using an optimization process to increase their complexity and explore their space beyond traditional models.

Contribution

It proposes a new model of complex networks that incorporates multiple topological heterogeneities and uses an optimization approach to enhance network complexity.

Findings

HC networks follow a distinct trajectory in principal component space during optimization.

Optimization leads to networks departing from classical models like Erdős-Rényi and Barabási-Albert.

Peripheral branching enhances network complexity after extensive optimization.

Abstract

The concept of 'complexity' plays a central role in complex network science. Traditionally, this term has been taken to express heterogeneity of the node degrees of a therefore complex network. However, given that the degree distribution is not enough to provide an invertible representation of a given network, additional complementary measurements of its topology are required in order to complement its characterization. In the present work, we aim at obtaining a new model of complex networks, called hypercomplex networks - HC, that are characterized by heterogeneity not only of the degree distribution, but also of a relatively complete set of complementary topological measurements including node degree, shortest path length, clustering coefficient, betweenness centrality, matching index, Laplacian eigenvalue and hierarchical node degree. The proposed model starts with uniformly random networks, namely Erdos-Renyi structures, and then applies optimization so as to increase an index of the overall complexity of the networks. An optimization approach has been considered in terms of gradient descent. The complexity index expresses the dispersion of the several considered measurements. Several interesting results are reported and discussed, including the fact that the HC networks define, as the optimization proceeds, a trajectory in the principal component space of the measurements that tends to depart from the considered theoretical models (Erdos-Renyi, Barabasi-Albert, Waxman, Random Geometric Graph and Watts-Strogatz), heading to a previously empty space (low density of cases). We observed that, after a considerably large number of optimization steps, peripheral branching tends to appear that further enhances the complexity of these networks.