Feature-specific prediction errors for visual mismatch.

Area of Science:

• Neuroscience
• Cognitive Science

Background:

• Predictive coding theory suggests the brain generates predictions about sensory input.
• A key untested prediction is that violated predictions generate distinct neural signals (precision weighted prediction errors - pwPEs) depending on the violated feature, even without attention.

Purpose of the Study:

• To test if distinct feature-specific predictions elicit distinct pwPEs.
• To investigate the neural basis of pwPEs for visual features like color and facial emotion, independent of attention.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used with a multi-feature visual mismatch paradigm.
• A computational model simulated pwPEs for a Bayes-optimal observer.
• Blood oxygen level dependent (BOLD) responses were analyzed while participants performed a distractor task, with eye-tracking controlling for attention.

Main Results:

• Unexpected color changes elicited pwPEs in the fusiform gyrus.
• Unexpected changes in facial emotion elicited pwPEs in cortico-thalamo-cerebellar regions linked to emotion and theory of mind.
• Predictions about emotions engaged fusiform, occipital, and temporal areas.

Conclusions:

• Results support a broad role for predictive coding across perceptual levels, from basic features to complex social cues.
• The findings suggest continuous, automatic monitoring of the social environment, irrespective of attentional state.