View-invariant object representation in anterior and posterior inferotemporal cortex: A machine learning approach

Area of Science:

• Neuroscience
• Computational Neuroscience
• Visual Perception

Background:

• The inferotemporal (IT) cortex is crucial for object recognition in the ventral visual stream.
• Previous studies indicated single cells in area TE exhibit viewing angle tolerance, unlike those in area TEO.

Purpose of the Study:

• To compare population activity between TE and TEO using machine learning to assess viewpoint tolerance.
• To investigate the representation of viewpoint invariance in different IT cortex subregions.

Main Methods:

• Utilized a machine learning algorithm (Support Vector Machine) to classify object views.
• Trained a classifier on object images at specific viewing angles and tested its performance on images at different angles.
• Analyzed population vectors derived from neural responses in areas TE and TEO.

Main Results:

• Area TE populations demonstrated viewing angle tolerance of 30-90° for objects with prior view-association learning.
• Area TEO populations showed limited viewing angle tolerance of 30°.
• A significant increase in performance (d') in area TE during the late time period suggests enhanced view-invariance representation.

Conclusions:

• Viewpoint invariance is more strongly expressed in the TE region of the inferotemporal cortex.
• The TEO region also contributes to viewpoint-invariant object recognition, albeit to a lesser extent.
• Late-period neural activity in TE may be critical for robust view-invariant object recognition.