Pairwise network information and nonlinear correlations.

摘要: Reconstructing the structural connectivity between interacting units from observed activity is a challenge across many different disciplines. The fundamental first step to establish whether or what extent interactions can be considered pairwise and, thus, modeled as an interaction network with simple links corresponding interactions. In principle, this determined by comparing maximum entropy given bivariate probability distributions true joint entropy. practical cases, not option since needed may reliably estimated optimization too computationally expensive. Here we present approach that allows one use mutual informations proxy for distributions. This has advantage of being less expensive and easier estimate. We achieve introducing novel maximization scheme based on conditioning entropies informations. renders our typically superior other methods linear approximations. advantages proposed method are documented using oscillator networks resting-state human brain generic relevant examples.