A unified model of the structure and function of primate visual cortex
Abstract/Contents
- Abstract
- Humans have the remarkable capacity to recognize visual objects despite challenging variations in their pose, illumination, and context. This ability depends on the ventral visual stream, a series of cortical areas that progressively transforms the signal from the retina into representations of object category, location, color, texture, and size. Our understanding of the function and development of the ventral visual stream is anchored in the tight coupling between structure and function in the constituent cortical areas: in each area, neurons are arranged in the cortical sheet according to the visual features they respond most strongly to. In the earliest stage of the ventral visual stream neighboring neurons preferentially respond to edges of similar orientations and colors, whereas neurons toward the end of the ventral stream cluster together according to their preferred object category, e.g., faces, limbs, and places. Understanding the development and purpose of this functional organization requires the construction of detailed models whose predictions can be evaluated against neural measurements. In this dissertation, I present topographic deep convolutional neural networks (topographic DCNNs) as unifying models of neural structure and function throughout the ventral visual stream. Topographic DCNNs implement the simple hypothesis that functional organization in the visual cortex can be reproduced by optimizing the parameters of a neural network to perform a challenging visual task while keeping local populations of neurons correlated with one another. I find that topographic DCNNs are able to reproduce functional organization in both early and later stages of the ventral visual stream, that this brain-model correspondence is strongest for more biologically-plausible learning algorithms, and that topographic DCNNs can be used to predict how changes to visual inputs during development will affect cortical map formation. The success of topographic DCNNs in the prediction of the functional organization of the primate ventral visual stream implies the existence of simple unifying principles for the development of those regions, and serves as a foundation from which increasingly accurate models of visual processing can be constructed.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2022; ©2022 |
| Publication date | 2022; 2022 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Margalit, Eshed |
|---|---|
| Degree supervisor | Grill-Spector, Kalanit |
| Degree supervisor | Yamins, Daniel |
| Thesis advisor | Grill-Spector, Kalanit |
| Thesis advisor | Yamins, Daniel |
| Thesis advisor | Druckmann, Shaul |
| Thesis advisor | Gardner, Justin, 1971- |
| Thesis advisor | Norcia, Anthony Matthew |
| Degree committee member | Druckmann, Shaul |
| Degree committee member | Gardner, Justin, 1971- |
| Degree committee member | Norcia, Anthony Matthew |
| Associated with | Stanford University, Neurosciences Program |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Eshed Margalit. |
|---|---|
| Note | Submitted to the Neurosciences Program. |
| Thesis | Thesis Ph.D. Stanford University 2022. |
| Location | https://purl.stanford.edu/pn433ry1536 |
Access conditions
- Copyright
- © 2022 by Eshed Margalit
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...