A common, high-dimensional model of the representational space in human ventral temporal cortex

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The human ventral temporal (VT) cortex processes complex visual information.
• Understanding the neural representations of visual stimuli across individuals is challenging due to inter-subject variability.
• Previous methods often struggle to align brain data effectively across different people.

Purpose of the Study:

• To develop a high-dimensional model of the representational space in the human VT cortex.
• To identify common response-tuning functions shared across individuals for representing visual stimuli.
• To introduce and validate a novel method, hyperalignment, for mapping individual brain data into a common representational space.

Main Methods:

• Utilized functional magnetic resonance imaging (fMRI) to measure brain activity patterns (response-pattern vectors).
• Developed and applied a novel technique called "hyperalignment" to map individual subject data into a shared model space.
• Used movie viewing data to identify common response-tuning functions and then tested these on category perception tasks.

Main Results:

• Identified 35 common response-tuning functions within the VT cortex that are consistent across individuals.
• Hyperalignment significantly improved between-subject classification (BSC) of neural response patterns compared to traditional anatomical alignment.
• The performance of classification in the common model space matched the performance of within-subject classification.

Conclusions:

• Population codes for complex visual stimuli in the human VT cortex are based on shared response-tuning functions across individuals.
• Hyperalignment provides a powerful tool for analyzing and comparing neural representations of visual information across subjects.
• This work advances our understanding of how the brain represents complex visual information in a standardized, cross-subject manner.