Neural surprise in somatosensory Bayesian learning.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The brain's ability to learn environmental statistical regularities is crucial for perception and behavior.
• Bayesian principles are thought to underlie how the brain models environmental dependencies.
• Specific algorithms for somatosensory learning and their neural correlates remain largely unexplored.

Purpose of the Study:

• To investigate the cortical dynamics of the somatosensory learning system.
• To determine the form of the generative model used in somatosensory learning.
• To identify neural signatures of surprise signaling during somatosensory processing.

Main Methods:

• Recorded electroencephalography (EEG) data from 40 participants during a somatosensory roving-stimulus paradigm.
• Applied single-trial modeling in sensor and source space across peri-stimulus time.
• Utilized Bayesian model selection to analyze evoked potentials and identify surprise encoding.

Main Results:

• Evoked potentials are best explained by a non-hierarchical Bayesian learning model employing leaky integration.
• Secondary somatosensory cortices encode confidence-corrected surprise around 70ms post-stimulus, indicating model inadequacy.
• Primary somatosensory cortex shows Bayesian surprise encoding from 140ms, suggesting model updating.

Conclusions:

• Early somatosensory processing reflects Bayesian perceptual learning, with distinct cortical areas handling different aspects of surprise.
• The dissociation between early and later surprise signals suggests a role in controlling model update rates.
• Findings contribute to understanding the neural mechanisms of Bayesian inference in somatosensory perception.