Inference of time-dependent causal influences in Networks

Abstract

We address the challenge of detecting time-variant interactions in multivariate systems. Inferring Granger-causal interactions between processes promises to gain deeper insights into mechanisms underlying network phenomena, e. g., in the neurosciences. Renormalized partial directed coherence (rPDC) has been introduced as a means to investigate Granger causality in such multivariate systems. When using rPDC a major challenge is the reliable estimation of parameters in vector autoregressive processes. For time-varying connections a time-resolved estimation of the coefficients is mandatory. We show that the State Space Model in combination with the Kalman filter is a powerful tool for estimating time-variate AR process parameters.