Building and evaluating computational models of the mammalian visual system
Abstract/Contents
- Abstract
- Animals continuously and dynamically process sensory information in service of both flexible and inflexible behaviours. To understand the brain's complex information-processing pipeline by which such behaviours arise, we must first understand how the brain transforms sensory information from its raw form. This will then allow us determine what information is accessible downstream in the process. In this dissertation, we try to understand how the brain processes visual information, which entails building and evaluating computational models that can predict how the animal will respond to novel visual inputs. We focus on a class of models known as convolutional neural networks (CNNs) and demonstrate ways in which they can be evaluated against and be built for primates and for rodents to better understand how the mammalian visual system supports behaviour. We first demonstrate a time-resolved correspondence between a feedforward CNN and whole-brain neural responses during human object processing and develop a data-driven optimization approach to improve upon correlations achieved between the model and the neural data. Motivated by extensive empirical work in rodents on navigational and on decision-making behaviours and by the desire to integrate models of cortical and of subcortical areas that support these behaviours, we build quantitatively accurate CNN models of the mouse visual system. Although CNNs are state-of-the-art models of primate and of rodent visual processing, they are extremely brittle. We therefore examine the nature of their brittleness and show the existence of representational differences between primary visual cortex of non-human primates and the models. Finally, we suggest that building less-brittle models will require us to incorporate the temporally-continuous nature of the visual inputs that animals receive. Looking forward, we hope that models of sensory cortex can be integrated with computational models of downstream cortical and subcortical areas, so that we can better understand how flexible and inflexible behaviours arise.
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 | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Kong, Nathan Cheuck Lam |
|---|---|
| Degree supervisor | Norcia, Anthony Matthew |
| Thesis advisor | Norcia, Anthony Matthew |
| Thesis advisor | Gardner, Justin, 1971- |
| Thesis advisor | Grill-Spector, Kalanit |
| Thesis advisor | Yamins, Daniel |
| Degree committee member | Gardner, Justin, 1971- |
| Degree committee member | Grill-Spector, Kalanit |
| Degree committee member | Yamins, Daniel |
| Associated with | Stanford University, School of Humanities and Sciences |
| Associated with | Stanford University, Department of Psychology |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Nathan C. L. Kong. |
|---|---|
| Note | Submitted to the Department of Psychology. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/qk198bh1136 |
Access conditions
- Copyright
- © 2023 by Nathan Cheuck Lam Kong
- 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...