Brown Digital Repository | Item

Full Metadata

Overview

Year:: 2022
Contributor:: Longoria-Valenzuela, Isabella (creator); Serre, Thomas (thesis advisor); Festa, Elena (reader); Brown University. Cognitive, Linguistic, and Psychological Sciences (sponsor)
Genre:: theses
Subject:: Cognitive neuroscience; Visual perception; Deep learning (Machine learning); Categorization (Psychology)
DOI: https://doi.org/10.26300/xmaw-7967

Files

Description

Abstract:: Our daily interaction with the visual world raises questions about how humans identify and categorize visual stimuli. Previous research on the Perceptual Magnet Effect (PME), which argues that a perceptual warping occurs at an ambiguous category boundary between category X and category Y, has focused on replicating this perceptual warping in other perceptual domains and using naturalistic visual stimuli. In the present study, GAN-generated stimuli were evaluated against human similarity judgments in a pairwise comparison similarity task. Specifically, the presented study used visual stimuli from the popular CIFAR-10 and FF-HQ datasets. By comparing average human similarity ratings during each interpolation step from category X to category Y, GAN-generated stimuli were measured on their ability to replicate the warping seen in the PME. The results suggest that the CIFAR-10 and FF-HQ visual stimuli show hints of perceptual warping within-class category and across-instance category boundaries, respectively, but additional experimental fine-tuning is needed to strengthen the results. These findings provide an important first step in using GAN-generated stimuli to replicate psychophysics experiments analyzing perceptual phenomena like the PME. In addition to improving the sample size and scaling up high-resolution visual stimuli, future work should aim to investigate the possible application of GAN-generated stimuli with modified latent spaces to other domains like memory and attention.
Notes:: Senior thesis (AB)--Brown University, 2022; Concentration: Cognitive Neuroscience

Content

Access Conditions

Use and Reproduction: All rights reserved
Rights: In Copyright
Restrictions on Use: All Rights Reserved

Citation

Longoria-Valenzuela, Isabella, "Revisiting Models of Visual Categorization Using Deep Learning Models: The Generation of Naturalistic Visual Stimuli Through GANs" (2022). Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations. Brown Digital Repository. Brown University Library. https://doi.org/10.26300/xmaw-7967

Relations

Collection:

Cognitive, Linguistic, and Psychological Sciences Theses and Dissertations

Theses and Dissertations for the Cognitive, Linguistic, and Psychological Sciences department.
...

Revisiting Models of Visual Categorization Using Deep Learning Models: The Generation of Naturalistic Visual Stimuli Through GANs