The neural code for "face cells" is not face-specific.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Primate Vision

Background:

• Face cells, neurons selectively responsive to faces, are primarily studied using facial stimuli.
• These neurons are clustered in the inferotemporal cortex, a region implicated in visual processing.
• Current research often assumes face selectivity arises from unique facial features.

Purpose of the Study:

• To investigate the nature of neural selectivity for faces by analyzing responses to a broad range of objects.
• To determine if non-face object properties predict face selectivity.
• To challenge the exclusive focus on face stimuli in understanding face-selective neurons.

Main Methods:

• Recorded neural responses in and around macaque face patches to hundreds of diverse objects.
• Analyzed response profiles of neurons to both face and non-face stimuli.
• Utilized deep neural networks trained on general object recognition to model neural responses.

Main Results:

• Found graded response profiles for non-face objects that correlated with the degree of face selectivity.
• Demonstrated that information encoded in general object-trained deep neural networks predicted neural responses better than color or simple shape.
• Showed that non-face object responses provided additional information about face-cell tuning.

Conclusions:

• Face selectivity in neurons is not solely based on face-specific features but on broader object representations.
• The tuning of category-selective neurons is better understood within a domain-general object space.
• Rethinking the methodology of studying category-selective neurons beyond their most effective stimuli is warranted.