Functional architectures in the brain. (360G-Wellcome-088130_Z_09_Z)

Our computational neuroscience programme has progressed at two levels: First, we have had considerable success in modelling observed brain responses using biologically plausible models of neuroimaging data. We can now model fMRI, EEG and MEG responses with neural mass models of distributed networks to estimate population dynamics, synaptic efficacy and receptor function. More importantly, we can compare different models using Bayesian model comparison to test competing hypotheses about underl ying functional architectures. The second line of work addresses the basic constraints under which open, self- organising biological systems like the brain operate. This work has led to a formulation (a free-energy principle), that places the Bayesian brain in the larger context of exchange with the environment. This formulation provides a unified account of action and perception and, within perception, perceptual inference, learning and attention. We would now like to bring these two lines o f work together and use neural mass models of empirical data to identify how the free-energy principle is instantiated in neuronal circuits. We have some clear hypotheses that are related formally to predictive coding theories of perception. We want to test these hypotheses using Bayesian model comparison and Dynamic Causal Models of electromagnetic brain responses.