A crucial aspect of this model is the relative contribution of bottom-up (prediction error) and top-down (prediction) influences on representational updates.
Perceptual inference – at a given level of the hierarchy – rests on the influence of prediction errors from lower levels, relative to the prediction error at the level in question.
In predictive coding, these influences are proportional to ‘precision’, which is an estimate of the signal-to-noise ratio or reliability of the prediction error (Feldman & Friston, 2010).
Physiologically, precision is thought to be encoded by the post-synaptic gain of the neurons that encode prediction error; namely, superficial pyramidal cells (Bastos et al., 2012 and Mumford, 1992).
If the gain of superficial pyramidal cells is relatively high in sensory areas, the propagation of sensory input (sensory prediction error) up the hierarchy is facilitated and top-down predictions are changed to match sensory input.
In this context, the percept is dominated by sensory input. On the other hand, if post-synaptic gain is relatively higher in upper levels, then top-down predictions are more precise and will dominate perceptual inference – being relatively impervious to imprecise bottom-up influences.