Data-driven analysis of interactions between pairs and ensembles of coupled dynamics
Petroula Laiou · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · 2017
The characterization of interactions between coupled dynamics from their signals is important for the understanding of real-world systems. The particular aspect of the detection of directional interactions has a central position in the analysis of dynamics. In simple unidirectionally coupled dynamics directional interactions can be achieved by applying data-driven approaches. However, for more complex dynamics the characterization of their directional interactions is not so straightforward. To address this problem we follow a data-driven approach by analyzing signals of pairs and ensembles of non-identical coupled dynamics. In particular, we use a nonlinear state-space approach and a phase-based approach. For the pairs of bidirectionally coupled dynamics, we introduce the notion of the coupling impact that allows us to better reveal the real effect that one dynamics has on the other for different degrees of asymmetry. Furthermore, we show that the coupling and its direction can be detected even for large ensembles of dynamics. Our results demonstrate that directional interactions in complex dynamics can be successfully inferred from the analysis of their signals. Hence, our work shows that the approaches are promising for a reliable detection of directional interactions from real-world signals.