The visually guided development of facial representations in the primate ventral visual pathway: A computer modeling...

Area of Science:

• Computational neuroscience
• Computer modeling of visual processing
• Primate ventral visual pathway research

Background:

• Neurons in the primate ventral visual pathway exhibit hierarchical processing of faces.
• Intermediate stages encode facial parts (eyes, nose), while later stages encode whole faces via spatial relations.

Purpose of the Study:

• To investigate the developmental mechanisms of face-selective neuronal firing properties.
• To model how unsupervised, visually guided learning shapes face perception in the ventral visual stream.

Main Methods:

• A hierarchical neural network model simulating the primate ventral visual pathway was developed.
• The model was trained using unsupervised learning with numerous random face stimuli.
• A local learning rule adjusted synaptic connection strengths between successive neuronal layers.

Main Results:

• The model successfully replicated experimentally observed neuronal firing properties for face processing.
• Separate representations for facial features (eyes, nose, mouth) emerged.
• Monotonically tuned representations of spatial relationships between features were developed.
• Independent representation of facial expression and identity was demonstrated.

Conclusions:

• Unsupervised learning in a hierarchical neural network can account for the development of face-selective neuronal properties.
• Neurons encoding global attributes likely represent specific spatial relationships between local features.
• The model provides insights into how the primate brain achieves complex visual recognition.