In this talk I will describe methods for making inferences about brain connectivity from imaging data (fMRI,MEG). Inferences are based on models of neural networks as described using differential equations. These models make various assumptions about how activity in one brain region changes activity in another, and how experimental tasks change the strength of connections. I will describe a model comparison criterion, based on the free energy, and show that it can accurately adjudicate between models. I will close with an example from an MEG study of working memory where we infer the mechanisms underlying large scale synchronization processes.
