Computer Science ETDs

Publication Date

Fall 11-15-2023

Abstract

Humans can leverage domain context to recognize novel patterns and categories based on limited known examples. In contrast, computational learning methods are not adept at exploiting context and require sufficient labeled examples to achieve similar accuracy. Many temporal data domain, for example, seismic signals and oil mining sensor data, requires domain expert annotation, which is both costly and time-consuming. The dependency on training data limits the applicability of machine learning algorithms for domains with limited labeled data. This dissertation aims to address this gap by developing temporal mining algorithms that exploit domain context to learn discriminative feature representation from limited samples to achieve improved performance. In this dissertation, I present four domain-specific feature representation learning methods for three diverse domains: a pattern detection algorithm for oil and gas mining, two classification methods for seismic activity monitoring, and a prediction model for user behavior in social media. We show empirical evidence of performance improvement using real datasets. We demonstrate these methods' practical usability for multiple real-world applications.

Language

English

Keywords

Temporal Data, Deep Learning, Feature Representation, Few-shot Learning

Document Type

Dissertation

Degree Name

Computer Science

Level of Degree

Doctoral

Department Name

Department of Computer Science

First Committee Member (Chair)

Abdullah Mueen

Second Committee Member

Manel Martinez-Ramon

Third Committee Member

Trilce Estrada

Fourth Committee Member

Bruna Jacobson

Download

Share

COinS