A unifying framework for functional organization in early and higher ventral visual cortex.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Primate Visual System Research

Background:

• Cortical systems exhibit functional organization, with neurons arranged in specific spatial patterns.
• The underlying principles governing the emergence of this functional organization in the cortex remain largely unknown.
• Understanding cortical functional organization is crucial for neuroscience and AI development.

Purpose of the Study:

• To develop a computational model predicting functional organization in primate visual cortical areas.
• To identify the key factors driving the emergence of functional organization in neural networks.
• To provide a unified principle for understanding the primate ventral visual system's organization.

Main Methods:

• Development of the topographic deep artificial neural network (TDANN).
• Analysis of TDANN's learning objectives: task-general sensory representation and spatial response smoothness.
• Evaluation of TDANN's representations against brain-like data and analysis of between-area connection lengths.

Main Results:

• The TDANN successfully predicts multiple aspects of functional organization in primate visual cortical areas.
• TDANN's success stems from balancing sensory representation learning with spatial smoothness, scaled by cortical area.
• TDANN-generated representations are more brain-like than those from unconstrained models, balancing performance and connection length.

Conclusions:

• The TDANN offers a predictive model for functional organization in the primate ventral visual system.
• A unified principle of balancing task learning with spatial smoothness explains observed cortical organization.
• This work bridges computational modeling and neuroscience, offering insights into brain development and AI architectures.