Feature learning for supervised knowledge discovery
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
In the expansive landscape of Artificial Intelligence (AI) / Machine Learning (ML), the identification and description of important data characteristics, termed features, is essential for solving complex real world problems. However, where do these features come from and how should they be used? The answer to these fundamental questions impact nearly every AI/ML approach and application. Traditionally, features have been either hand-crafted by human experts or data-driven derived through AI/ML. Herein, the notion of data driven features or, more specifically, feature learning is thoroughly investigated, inside of this dissertation, in the context of adversarial learning, open set recognition, and self supervised learning. First, I demonstrate a feature learning framework, driven by evolutionary optimization, that enables knowledge discovery with respect to adversarial learning. Second, I demonstrate an alternative feature learning framework, driven by neural metric learning and self supervision, that enables knowledge discovery with respect to open set recognition. Collectively, the capabilities of these frameworks are demonstrated on computer vision applications. Experiments illustrate domain improvements, including improved accuracy on open set recognition benchmarks and the capacity to discover meaningful patterns within unknown datasets.
Table of Contents
DOI
PubMed ID
Degree
Ph. D.