Date of Award

8-23-2024

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Psychology

Advisor(s)

Michael Kalish

Keywords

Category Learning;Computational Modeling;Generalized Context Model;Semi-Supervised Learning

Subject Categories

Cognitive Psychology | Psychology | Social and Behavioral Sciences

Abstract

Category learning, a fundamental aspect of human cognition, enables individuals to organize and interpret the vast array of stimuli in their environment. While traditionally studied under supervised and unsupervised learning frameworks, semi-supervised learning – combining small amounts of labeled data with larger pools of unlabeled data – has emerged as a promising area of research. This study investigates the mechanisms of human semi-supervised category learning, addressing the underexplored question of how unlabeled instances influence the learning process. Motivated by inconsistent findings in prior research, I explore the types of information people extract from unlabeled instances and whether this information can influence category learning processes. The experiments tested the hypothesis that human learners integrate information from unlabeled item distributions to shift their dimensional attention during category learning, offering empirical evidence for human semi-supervised learning and highlighting the importance of distributional characteristics – particularly, dimensional variability – in attention reallocation during semi-supervised learning. To quantify these attentional shifts, I propose a novel tool for measuring dimensional attention with fewer assumptions: the Geometric Dimensional Attention Estimator (GDE). Simulations comparing the GDE with the Generalized Context Model (GCM) highlight its strengths and limitations. Finally, I conclude with a broader discussion and summary of the findings, along with recommendations for future research.

Access

Open Access

Share

COinS