Date of Award

Spring 1-1-2018

Document Type

Thesis

Degree Name

Master of Arts (MA)

First Advisor

Randy C. O'Reilly

Second Advisor

Tim Curran

Third Advisor

Matt Jones

Abstract

Understanding the relationship between sets of objects is a fundamental requirement of cognitive skills, such as learning from example or generalization. For example, recognizing that planets revolve around stars, and not the other way around, is essential for understanding astronomical systems. However, the method by which we recognize and apply such relations is not clearly understood. In particular, how a set of neurons is able to represent which object fulfills which role (role binding), presented difficulty in past studies. Here, we propose a systems-level model, which utilizes selective attention and working memory, to address issues of role binding. In our model, selective attention is used to perceive visual stimuli such that all relations can be reframed as an operation from one object unto another, and so binding becomes an issue only in the initial recognition of the direction of the relation. We test and refine this model, utilizing EEG during a second-order relational reasoning task. Epoched EEG was projected to the cortical surface, providing sourcespace estimates of event related potentials. Permutation testing revealed 8 cortical clusters which responded differentially based on the specifics of a trial. Dynamic connectivity between these clusters was estimated with the directed transfer function, to reveal the dynamic causality between regions. Our results support the model, identifying a distinct bottom-up network that identifies relations between single pairs of objects, along with a top-down biasing network that may reorient attention to sequential pairs of objects. Taken together, our results show that relational reasoning can be performed by a distributed network, utilizing selective attention

Share

COinS