4.3.2. Neuronal model of MH
Our model of MH is based on the ‘predictive coding’ theory of brain function (Bastos et al., 2012, Kumar et al., 2011 and Rao and Ballard, 1999).
In this framework, each level of the cortical hierarchy tries to predict the representation of sensory objects in the level below by sending top-down predictions.
Aspects of the representation that are inconsistent with the prediction (the prediction error) are then passed back to the higher level.
Prediction errors are then used to update the representations at the higher level.
In this framework, all bottom-up (ascending) connections communicate prediction error, and top-down (descending) connections convey predictions.
This message passing changes hierarchical representations such that prediction error is minimized at all levels.
In this regard, the predictive coding framework is Bayes-optimal from the perceptual inference perspective ( Friston, 2010).