Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Jorge_Chang_Thesis_PreDefenseDraft.pdf (19.24 MB)

ETD Abstract Container

Abstract Header

Probing Human Category Structures with Synthetic Photorealistic Stimuli

Chang Cheng, Jorge
http://orcid.org/0000-0002-9289-4480
http://rave.ohiolink.edu/etdc/view?acc_num=osu1658321352580774

Abstract Details

2022, Doctor of Philosophy, Ohio State University, Psychology.
Formal theories of human categorization have been a focus of research in psychology and bolster more than a century's worth of studies. In this process, many theories have been proposed, evaluated, and compared through the development of computational models. An assumption shared among these studies is the existence of a psychological stimulus space that encodes stimuli. Traditionally, the field has modeled this space through low-dimensional artificial stimuli designed for the laboratory setting. While this paradigm grants a high degree of control over experiments, the ecological validity of this assumption is a major point of criticism. After all, stimuli in the natural world are often complex and unlikely to be fully represented in a low-dimensional space. Thus, studies in the field must address this representational gap for results to be generalizable. In the present thesis, we study this representational gap by reexamining the debate between exemplar and prototype models of categorization in an experiment that utilizes highly realistic stimuli and a high-dimensional stimulus space. In doing so, we propose a framework for experimentation based on using Generative Adversarial Networks (GANs) to model the psychological feature space. This framework also showcases several techniques that address the unique challenges of utilizing high-dimensional stimuli that have prevented their adoption in the past. We employed our framework in two experiments comparing prototype and exemplar models in different settings and found a consistent advantage for prototype models contrary to the dominant view in the field. We then theorize an explanation for this advantage by discussing the effects of increasing the dimensionality of the feature space on each type of model, arguing that prototype models are more robust in these scenarios. These observations suggest that previously found advantages for exemplar models might have been an artifact of using low-dimensional stimuli. The results in this thesis represent a first step towards uncovering the underlying structure of natural categories and highlight the importance of ecological validity in psychological experiments.
Jay Myung Myung (Advisor)
Keith Redmill (Committee Member)
Brandon Turner (Committee Member)
Mark Pitt (Advisor)
146 p.
Cognitive Psychology; Computer Science; Psychobiology
Categorization, GAN, DNN, Bayesian Optimization, Optimal Experimental Design, Prototype, Exemplar, Ecological Validity, High-dimensional spaces, Model comparison,

Recommended Citations

Citations

  • Chang Cheng, J. (2022). Probing Human Category Structures with Synthetic Photorealistic Stimuli [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1658321352580774

    APA Style (7th edition)

  • Chang Cheng, Jorge. Probing Human Category Structures with Synthetic Photorealistic Stimuli. 2022. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1658321352580774.

    MLA Style (8th edition)

  • Chang Cheng, Jorge. "Probing Human Category Structures with Synthetic Photorealistic Stimuli." Doctoral dissertation, Ohio State University, 2022. http://rave.ohiolink.edu/etdc/view?acc_num=osu1658321352580774

    Chicago Manual of Style (17th edition)

osu1658321352580774
89
© 2022, some rights reserved.Creative Commons License Probing Human Category Structures with Synthetic Photorealistic Stimuli by Jorge Chang Cheng is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by The Ohio State University and OhioLINK.