Electronic Theses and Dissertations

Identifier

2649

Date

2016

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Engineering

Concentration

Computer Engineering

Committee Chair

Mohammed Yeasin

Committee Member

Gavin M Bidelman

Committee Member

David Kimbrough Oller

Committee Member

Russel Jerry Deaton

Abstract

Working memory (WM) is considered a core element of cognition, acting as a shared resource among many different mental processes. It plays a key role in determining individual cognitive capacity and performance limits. Despite remarkable progress, prior works fall short in identifying which network structure in the brain limits our working memory capacity. We investigated this issue by analyzing electrical activity of human brain recorded from the scalp (Electroencephalogram or EEG). In this dissertation, first we describe a novel method for constructing a graph model of human brain activities. The nodes and edges of the graph represent cortices and interconnecting links, respectively. Next, we present a multivariate machine learning approach for identifying cognitive states and estimating network structure explaining individual differences. Finally, we describe our proposed approach that preserves the spatial-spectral-temporal information and capable of learning representation from EEG recording eliminating the need of feature engineering and reliance on domain knowledge.Main contributions of this dissertation are (but are not limited to): 1) study and characterization of common oscillatory neural response to varying levels of memory demand (load); 2) Identification of singular cortical structure explaining differences in WM capacity across individuals; and 3) Development of a robust representation learning approach for multi-dimensional time-series data. Our findings on identifying commonality and singularity in cortical structures are critical for designing cognitively informed brain-computer interfaces and long-term goal of designing neural prosthetics. On the other hand, our proposed representation learning method opens up new possibilities for data driven decoding of brain activities which is particularly capable in generalizing across individuals.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.

Share

COinS