Measuring instantaneous directed dependencies in interacting oscillators
Bruno Bauwens, Bart Wyns, Dieter Devlaminck, Georges Otte, Luc Boullart, Patrick Santens · Ghent University Academic Bibliography (Ghent University) · 2007
Algorithmic information transfer has been theoretically shown to detect directed dependency or causality in bivariate signals in an optimal way.Here its practical use in a non-ideal setting is investigated.First we show the close connection to the common tests for directed interactions.Subsequently, consequences of model non-ideality is described and a deep connection between regularization and (directed) independence tests is concluded.Thereafter, a directed independence test is constructed to detect the presence of phase coupling between two oscillators using support vector regression.Finally the algorithm is used to determine moments of interaction for dynamic-coupled harmonic oscillators, by exploiting full length information of the signals.