Modelling and Measuring the Predictive Brain

The Free Energy Principle (FEP) has been proposed as a unifying explanation of adaptive behaviour within self-organising systems subsuming perception, action, and cognition under one computational umbrella. Within neuroscience, the FEP serves as the mathematical and normative foundation for two influential models: predictive coding and active inference, which together generate testable predictions across the aforementioned domains. Still, with explanatory ambitions as grand as the FEP’s there remain many areas that are either empirically under explored, or are too theoretically under developed to be explored empirically. As such this thesis has two complementary aims. The first aim is to test a set of the predictions of the classic formulation of predictive coding by examining the temporal effects of prediction error at multiple levels of the visual hierarchy using time-resolved multivariate pattern analysis and electroencephalography. The second aim is primarily theoretical. As it stands it is not clear how, under active inference or predictive coding, perceptual information transitions from unconscious processing, where it is inaccessible from the first person point of view, to conscious processing where it is available for subjective report. To help bridge this gap, we will present a computational model of conscious access based on the partially observable Markov decision process formulation of active inference. Through simulation we demonstrate the model’s ability to: i) reproduce a wide range of electrophysiological and behavioural results; ii) reconcile/explain (apparently) conflicting findings; and iii), generate novel predictions that can be tested empirically. In doing so we aim to expand the explanatory scope of the FEP to experimental paradigms commonly used in consciousness science