Fourier Bayesian Information Criterion for network structure and causality estimation

We propose a variant of the Bayesian Information Criterion (BIC) for network structure learning that we have called Fourier BIC (FBIC). The new measure is based on spectral techniques and can be applied in a similar way to previous network fitting measures such as Akaike's, Minimum description length or BIC. FBIC presents the advantage of causality estimation, which is of paramount importance in dynamic networks and complex systems analysis. We test the performance of FBIC by estimating the structure of a causal Gaussian network using the K2 algorithm.