Pre-frontal cortex guides dimension-reducing transformations in the occipito-ventral pathway for categorization...

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The brain's visual categorization relies on hierarchical computations transforming high-dimensional inputs into lower-dimensional representations (manifolds).
• Understanding these transformations is crucial for explaining diverse categorization behaviors.

Purpose of the Study:

• To investigate the neural mechanisms underlying visual categorization.
• To test the hypothesis of hierarchical transformations supporting multiple categorization tasks.

Main Methods:

• Magnetoencephalography (MEG) source activity analysis.
• Participants performed multiple categorization tasks (face expression, face gender, pedestrian gender, vehicle type) on the same stimuli.
• Analysis of dynamic brain activity across different time windows (50-350 ms).

Main Results:

• Three distinct transformation stages were identified, guided by the pre-frontal cortex.
• Stage 1 (50-120 ms): Occipital sources encode task-relevant and irrelevant features; relevant features move to ventral/dorsal regions, irrelevant features halt.
• Stage 2 (121-150 ms): Feature representations reduce to lower-dimensional manifolds.
• Stage 3 (161-350 ms): Manifolds transform into task-relevant features for categorization.

Conclusions:

• The brain employs a hierarchical network mechanism to process visual information.
• This process transforms high-dimensional inputs into specific feature manifolds supporting flexible categorization.
• Findings elucidate how the brain achieves multiple categorization behaviors from complex visual stimuli.