Within-category representational stability through the lens of manipulable objects

Area of Science:

• Cognitive Neuroscience
• Neuroimaging
• Computer Vision

Background:

• Human object recognition is a key cognitive ability, distinguishing us from other species and AI.
• Understanding how object representations remain stable across time and contexts is crucial but poorly understood.
• Investigating the neural basis of stable within-category object representations is essential.

Purpose of the Study:

• To investigate the neural stability of within-category object representations across different scanning days.
• To identify brain regions supporting stable object representations.
• To determine which types of features (e.g., visual, semantic) best explain this neural stability.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to measure brain activity.
• Representational similarity analysis was applied to activity patterns from 80 tool images scanned on separate days.
• Representational geometries were compared with those from convolutional neural network layers and visual similarity models.

Main Results:

• Within-category representational stability was found in temporal, parietal, and premotor cortex.
• These stable regions showed a stronger relationship with higher-level visual and semantic features.
• Neural stability correlated more with perceived and abstract visual information than veridical, lower-level features.

Conclusions:

• Neural stability of within-category object representations is observable in tool-processing brain regions.
• This stability is likely driven by abstract, higher-level visual and semantic information.
• Stable object representations may arise from cross-talk between category-specific brain regions.