Impact of lag information on network inference

TL;DR

This paper demonstrates that lag information from cross-correlation analysis can improve the inference of network connectivity in complex systems, using experimental data from coupled chaotic oscillators.

Contribution

It introduces the use of lag times from cross-correlation to enhance network inference accuracy in complex systems.

Findings

Lag times provide valuable information about true network connections.

Cross-correlation lag analysis improves inference accuracy.

Experimental validation with chaotic oscillators supports the method.

Abstract

Extracting useful information from data is a fundamental challenge across disciplines as diverse as climate, neuroscience, genetics, and ecology. In the era of ``big data'', data is ubiquitous, but appropriated methods are needed for gaining reliable information from the data. In this work we consider a complex system, composed by interacting units, and aim at inferring which elements influence each other, directly from the observed data. The only assumption about the structure of the system is that it can be modeled by a network composed by a set of $N$ units connected with $L$ un-weighted and un-directed links, however, the structure of the connections is not known. In this situation the inference of the underlying network is usually done by using interdependency measures, computed from the output signals of the units. We show, using experimental data recorded from randomly coupled electronic R{\"o}ssler chaotic oscillators, that the information of the lag times obtained from bivariate cross-correlation analysis can be useful to gain information about the real connectivity of the system.