Representation of multiple objects in macaque category-selective areas.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Perception

Background:

• Object recognition typically occurs amidst multiple objects, yet neural responses are often studied using isolated stimuli.
• The inferotemporal cortex (IT) is crucial for object recognition, containing specialized regions like the middle lateral face patch (ML) and middle body patch (MB).

Purpose of the Study:

• To investigate the rules governing neural responses in IT cortex when presented with multiple objects.
• To determine if existing normalization frameworks can explain responses to complex visual scenes.

Main Methods:

• Recording neural activity from single cells in the ML and MB patches of macaque IT cortex.
• Presenting pairs of objects to elicit responses and analyzing these responses within a normalization framework.

Main Results:

• Neural responses to object pairs in ML and MB cells are explained by normalization, with an added factor of homogeneous category selectivity in the normalization pool.
• This model predicts diverse behaviors including winner-take-all, contralateral-take-all, and weighted averaging, contingent on object category, spatial arrangement, and contrast.
• Winner-take-all dynamics were observed, suggesting a mechanism for clutter-invariant representation of faces and bodies.

Conclusions:

• Normalization, augmented by category-selective neural neighborhoods, effectively models IT cortex responses to multiple objects.
• This framework provides insights into how the brain achieves robust object recognition in natural, cluttered environments.
• The findings highlight a potential neural mechanism for maintaining object representations despite visual clutter.