| Summary: | Estimation of uncertainty is an important aspect of perception and a prerequisite for effective action. This thesis explores the implementation of uncertainty estimation as precision modulation within a predictive coding hierarchy, optimised within a neurbiologically-plausible message-passing scheme via the minimisation of free-energy. This thesis consists of six chapters. The first presents a new model of a classic visual illusion, the Cornsweet illusion, which demonstrates that the Cornsweet illusion is a natural consequence of Bayes-optimal perception under the free-energy principle, and demonstrates that increasing contrast can be modelled by increasing signal-to-noise ratio. The second chapter describes dynamic causal modelling of EEG data collected from participants viewing the Cornsweet illusion, demonstrating that a reduction in precision, or superficial pyramidal cell gain, in lower visual hierarchical levels, is sufficient to explain contrast-dependent changes in ERPs. The third describes a model of a simple attentional paradigm – the Posner paradigm – recasting attention as the optimal modulation of precision in sensory channels. The fourth describes an MEG study of the Posner paradigm, using Bayesian model selection to explore the role of changes in backwards and modulatory connections and changes in local superficial pyramidal cell gain in producing the electrophysiological and behavioural correlates of the Posner paradigm. The fifth chapter recasts the Posner paradigm in the motor domain to investigate the level (intrinsic vs. extrinsic) of precision modulation by motor cues. The sixth describes a new model of sensory attenuation based on using precision modulation to balance the imperatives to act and perceive. I hope to demonstrate that precision modulation within predictive coding hierarchies, under the free-energy principle, is a flexible and powerful way of describing and explaining both behavioural and neuroimaging data.
|
|---|
