Classic approaches to functional connectivity of two brain regions consider the correlation between regional average time courses. This type of analysis makes it possible to infer whether two regions are co-active or not. Another level of inference is to test whether two regions have shared information. For that, multi-dimensional connectivity methods are suited. Those methods do not require averaging, that can lose information, and target the information content of high-dimensional activity patterns. In my presentation I will talk about a class of multi-dimensional connectivity methods, named representational connectivity analysis. I will talk about different variants of RCA (model-free and model-based RCA) and the different types of inferences they support. Finally, I will talk about a geometric-aware extension of model-free RCA, distance between 2nd moment matrices on the Riemannian manifold, which makes it possible to test for co-activation and shared information simultaneously.
Postdoctoral Researcher @ UKE Hamburg