Detecting spatio-temporal modes in multivariate data by entropy field decomposition (Q2826727)

The goal in this paper is to develop a method for the investigation of spatio-temporal signal fluctuations in time-resolved noisy volumetric data acquired from nonlinear and non-Gaussian systems. The authors intend to develop a theoretical framework based on the information field theory (IFT) and the entropy spectrum pathways (ESP) which can be extended to a broad range of problems. The new method is called entropy field decomposition (EFD) and its purpose is to detect and separate spatio-temporal modes that are non-Gaussian and nonlinear in space and time variable and are produced by complex spatio-temporal interacting pathways. The EFD should allow an efficient characterization and ranking of modes within a complex multivariate spatio-temporal data. The IFT, the ESP approaches and then the EFD formalism together with computational aspects are presented. The new method is implemented on two examples. In Appendices A and B, the FMRI and TOMADO data are explained.