You are here

Categorical Change: Exploring the Effects of Concept Drift in Human Perceptual Category Learning

Download pdf | Full Screen View

Date Issued:
2018
Abstract/Description:
Categorization is an essential survival skill that we engage in daily. A multitude of behavioral and neuropsychological evidence support the existence of multiple learning systems involved in category learning. COmpetition between Verbal and Implicit Systems (COVIS) theory provides a neuropsychological basis for the existence of an explicit and implicit learning system involved in the learning of category rules. COVIS provides a convincing account of asymptotic performance in human category learning. However, COVIS (-) and virtually all current theories of category learning (-) focus solely on categories and decision environments that remain stationary over time. However, our environment is dynamic, and we often need to adapt our decision making to account for environmental or categorical changes. Machine learning addresses this significant challenge through what is termed concept drift. Concept drift occurs any time a data distribution changes over time. This dissertation draws from two key characteristics of concept drift in machine learning known to impact the performance of learning models, and in-so-doing provides the first systematic exploration of concept drift (i.e., categorical change) in human perceptual category learning. Four experiments, each including one key change parameter (category base-rates, payoffs, or category structure [RB/II]), investigated the effect of rate of change (abrupt, gradual) and awareness of change (foretold or not) on decision criterion adaptation. Critically, Experiments 3 and 4 evaluated differences in categorical adaptation within explicit and implicit category learning tasks to determine if rate and awareness of change moderated any learning system differences. The results of these experiments inform current category learning theory and provide information for machine learning models of decision support in non-stationary environments.
Title: Categorical Change: Exploring the Effects of Concept Drift in Human Perceptual Category Learning.
12 views
7 downloads
Name(s): Wismer, Andrew, Author
Bohil, Corey, Committee Chair
Szalma, James, Committee Member
Neider, Mark, Committee Member
Gluck, Kevin, Committee Member
University of Central Florida, Degree Grantor
Type of Resource: text
Date Issued: 2018
Publisher: University of Central Florida
Language(s): English
Abstract/Description: Categorization is an essential survival skill that we engage in daily. A multitude of behavioral and neuropsychological evidence support the existence of multiple learning systems involved in category learning. COmpetition between Verbal and Implicit Systems (COVIS) theory provides a neuropsychological basis for the existence of an explicit and implicit learning system involved in the learning of category rules. COVIS provides a convincing account of asymptotic performance in human category learning. However, COVIS (-) and virtually all current theories of category learning (-) focus solely on categories and decision environments that remain stationary over time. However, our environment is dynamic, and we often need to adapt our decision making to account for environmental or categorical changes. Machine learning addresses this significant challenge through what is termed concept drift. Concept drift occurs any time a data distribution changes over time. This dissertation draws from two key characteristics of concept drift in machine learning known to impact the performance of learning models, and in-so-doing provides the first systematic exploration of concept drift (i.e., categorical change) in human perceptual category learning. Four experiments, each including one key change parameter (category base-rates, payoffs, or category structure [RB/II]), investigated the effect of rate of change (abrupt, gradual) and awareness of change (foretold or not) on decision criterion adaptation. Critically, Experiments 3 and 4 evaluated differences in categorical adaptation within explicit and implicit category learning tasks to determine if rate and awareness of change moderated any learning system differences. The results of these experiments inform current category learning theory and provide information for machine learning models of decision support in non-stationary environments.
Identifier: CFE0007114 (IID), ucf:51947 (fedora)
Note(s): 2018-05-01
Ph.D.
Sciences, Psychology
Doctoral
This record was generated from author submitted information.
Subject(s): category learning -- concept drift -- COVIS -- base-rates -- payoffs -- multiple systems -- nonstationary
Persistent Link to This Record: http://purl.flvc.org/ucf/fd/CFE0007114
Restrictions on Access: public 2018-05-15
Host Institution: UCF

In Collections