Neural Representations of Ensemble Mean and Variance Across Visual Features

Area of Science:

• Cognitive Neuroscience
• Neuroimaging
• Visual Perception

Background:

• Humans efficiently extract summary statistics (e.g., mean, variance) from complex visual scenes, crucial for visual cognition.
• Unresolved questions concern the neural basis of ensemble perception: shared vs. feature-specific systems and shared vs. dissociable mechanisms for different statistics.

Purpose of the Study:

• To investigate the neural representation of ensemble mean and variance across different visual features (orientation, shape, animacy).
• To determine if ensemble statistics rely on common or distinct neural systems and mechanisms.

Main Methods:

• Functional magnetic resonance imaging (fMRI) and multivariate pattern analysis (MVPA) were employed.
• Whole-brain searchlight and region-of-interest (ROI) analyses examined neural decoding of ensemble statistics.
• Data were collected from 24 participants over two fMRI sessions.

Main Results:

• Ensemble mean and variance information is distributed across the visual cortex but differentially weighted.
• Mean decoding was stronger in ventral visual regions, showing a posterior-anterior gradient and largely feature-specific organization.
• Variance decoding was stronger in dorsal parietal and frontoparietal regions, generalizing across features and showing overlap.

Conclusions:

• A graded division of labor exists between ventral and dorsal visual pathways for processing ensemble statistics.
• Feature-specific and feature-general neural codes coexist within the visual cortex for ensemble perception.
• Findings reconcile conflicting evidence regarding the neural mechanisms and specificity of ensemble representations.