Equivalent processing of facial expression and identity by macaque visual system and task-optimized neural network.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Primate visual systems and deep neural networks (DNNs) excel at classifying facial expressions and identities simultaneously.
• The underlying neural computations for these tasks in both systems remain poorly understood.

Purpose of the Study:

• To investigate the neural computations shared and segregated between primate visual systems and DNN models for facial expression and identity classification.
• To compare fMRI representations in the macaque visual cortex with a multi-task DNN model.

Main Methods:

• Developed a multi-task DNN model for classifying monkey facial expressions and identities.
• Compared fMRI neural representations from macaque visual cortex with the DNN model's representations.
• Utilized correspondence analysis to map DNN layers to specific macaque visual areas.

Main Results:

• Both systems share early stages for low-level face feature processing.
• Later stages segregate into distinct branches for facial expression and identity.
• Increased specificity for either expression or identity processing occurs in higher stages.
• The amygdala and anterior fundus (AF) face patch align with the DNN's expression branch; the anterior medial (AM) face patch aligns with the DNN's identity branch.

Conclusions:

• The macaque visual system and DNN models exhibit significant anatomical and functional similarities.
• These similarities suggest common computational mechanisms for face processing in biological and artificial systems.