Abstract

Complex network analysis emerges as a wide-spread analytical tool in many domains of science. It assumes that we can observe both the entities that represent the network nodes as well as the relations among them. In numerous scenarios, where the relations are not always immediately observable, we encounter the task of network inference. Given the observation of the dynamical change of the nodes’ properties, the task is to reconstruct the unobservable links among the nodes. The talk will present three approaches to network inference that build upon statistical measures of pairwise associations between nodes, machine learning methods for feature ranking, and equation discovery methods for modeling network dynamics. We will also consider brief comparative analysis of the approaches, illustrate their utility on tasks of reconstructing known networks and discuss the prospects for developing an integrative approach.

• Kuzmanovski V, Todorovski L, Dzeroski S (2018) GigaScience 7(11): giy118. doi:10.1093/gigascience/giy118
• Leguia MG, Levnajic Z, Todorovski L, Zenko B (2019) Chaos 29: 093107. doi:10.1063/1.5092170 also arXiv:1902.03896
• Simidjievski N, Tanevski J, Zenko B, Levnajic , Todorovski L, Dzeroski S (2018) New Journal of Physics 20(11): 113003. doi:10.1088/1367-2630/aae941 also arXiv:1712.03100