Abstract
Brain function can be conceived as a hierarchy of generative models that optimizes predictions of sensory inputs and minimizes "surprise." Each level of the hierarchy makes predictions of neural events at a lower level in the hierarchy, which returns a prediction error when these expectations are violated. We tested the generalization of this hypothesis to multiple sequential deviations, and we identified the most likely organization of the network that accommodates deviations in temporal structure of stimuli. Magnetoencephalography of healthy human participants during an auditory paradigm identified prediction error responses in bilateral primary auditory cortex, superior temporal gyrus, and lateral prefrontal cortex for deviation by frequency, intensity, location, duration, and silent gap. We examined the connectivity between cortical sources using a set of 21 generative models that embedded alternate hypotheses of frontotemporal network dynamics. Bayesian model selection provided evidence for two new features of functional network organization. First, an expectancy signal provided input to the prefrontal cortex bilaterally, related to the temporal structure of stimuli. Second, there are functionally significant lateral connections between superior temporal and/or prefrontal cortex. The results support a predictive coding hypothesis but go beyond previous work in demonstrating the generalization to multiple concurrent stimulus dimensions and the evidence for a temporal expectancy input at the higher level of the frontotemporal hierarchy. We propose that this framework for studying the brain's response to unexpected events is not limited to simple sensory tasks but may also apply to the neurocognitive mechanisms of higher cognitive functions and their disorders.
Highlights
Brain function can be conceived as a hierarchy of generative models that optimizes predictions of sensory inputs and prediction errors (Friston and Kiebel, 2009)
Brain function can be conceived as a hierarchy of generative models that optimizes predictions of sensory inputs and minimizes “surprise.” Each level of the hierarchy makes predictions of neural events at a lower level in the hierarchy, which returns a prediction error when these expectations are violated
We examined the connectivity between cortical sources using a set of 21 generative models that embedded alternate hypotheses of frontotemporal network dynamics
Summary
Brain function can be conceived as a hierarchy of generative models that optimizes predictions of sensory inputs and prediction errors (Friston and Kiebel, 2009). Under this generalized prediction hypothesis, top-down predictions are compared with bottom-up sensory inputs and return prediction errors when unexpected stimuli occur (Rao and Ballard, 1999; Kiebel et al, 2008; Friston, 2009; Chennu et al, 2013; Lieder et al, 2013b).
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have