Network dynamics reconstruction from data

Abstract

We consider the problem of recovering the model of a complex network of interacting dynamical units from time series of observations. We focus on typical networks which exhibit heterogeneous degrees, i.e. where the number of connections varies widely across the network, and the coupling strength for a single interaction is small. In these networks, the behavior of each unit varies according to their connectivity. Under these mild assumptions, our method provides an effective network reconstruction of the network dynamics. The method is robust to a certain size of noise and only requires relatively short time series on the state variable of most nodes to determine: how well-connected a particular node is, the distribution of the nodes' degrees in the network, and the underlying dynamics.