A computational account of topographic organization in the high-level visual cortex of primates



We introduce the Interactive Topographic Network (ITN), a computational framework for modeling the cortical organization of high-level vision. Through simulations of ITN models, we demonstrate that the topographic clustering of domains in the inferotemporal cortex of primates can result from the demands of visual recognition under biological constraints on wiring cost and the modulating sign of neural connections. The model’s learned organization is highly specialized but not entirely modular, capturing many organizational properties in higher-order primates. Our work is important for cognitive neuroscience, by providing a general developmental account of the field of topographic functional specialization, and for computational neuroscience, by demonstrating how well-known biological details can be incorporated into neural network models to account for empirical results.


The inferotemporal (IT) cortex in humans and other primates is topographically organized, containing multiple hierarchically organized areas selective for particular domains, such as faces and scenes. This organization is usually thought of in terms of domain-specific evolved visual mechanisms. Here, we develop an alternative, general and developmental domain account of computational cortical organization. The count is instantiated in Interactive Topographic Networks (ITNs), a class of computational models in which a hierarchy of model computational areas, subject to biologically plausible connectivity constraints, learns high-level visual representations optimized for multiple domains. We find that minimizing a wiring cost on the spatially organized feedforward and lateral connections, alongside realistic constraints on the sign of neural connectivity within the computational model, results in a hierarchical topographic organization. This organization replicates a number of key properties of the primate computational cortex, including the presence of domain-selective spatial clusters preferentially involved in the representation of faces, objects, and scenes; columnar responses through separate excitatory and inhibitory units; and a generic spatial organization in which the response correlation of pairs of units decreases with their distance. We therefore argue that topographic domain selectivity is an emergent property of a visual system optimized to maximize behavioral performance under generic connectivity-based constraints.


    • Accepted November 30, 2021.
  • Author contributions: research designed by NMB, MB and DCP; NMB has done research; NMB provided new analytical reagents/tools; NMB, MB and DCP analyzed the data; and NMB, MB and DCP wrote the article.

  • Assessors: MA, University of Pennsylvania; TK, Radboud University; and PS, Massachusetts Institute of Technology.

  • The authors declare no competing interests.

  • This article contains additional information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.2112566119/-/DCSupplemental.


About Author

Comments are closed.