Facilitating Visual to Parametric Interaction with Deep Contrastive Learning

Date
2020
Authors
Wooton, Billy Troy
Contributor
Advisor
Leigh, Jason
Department
Computer Science
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
This thesis presents an approach to facilitating Visual to Parametric Interactions (V2PIs) by leveraging deep contrastive learning models. Within the larger contexts of Human-in-the-loop Machine Learning and interactive visual analytics, V2PI systems aim to empower domain scientists and analysts to manipulate the internal parameters of a parametric projection algorithm via intuitive interactions with a 2D visualization of data. This visualization is generated by the projection algorithm, whose internal parameters dictate how highly dimensional input data are projected down to two-dimensions. In a V2PI system, the domain expert then interacts with this visualization directly, re-positioning points within the 2D projection space based on their domain knowledge and intuition, with the goal of not only exploring alternative projections, but also teaching the parametric algorithm to extract their domain knowledge, and apply it to new out-of-sample data points. In recent years, deep contrastive learning models have risen as a powerful way to learn desirable projections of high-dimensional data, such that data points similar to one another are tightly clustered within the 2D projection, while dissimilar points are spread apart. This thesis explores deep contrastive learning as a compelling candidate for use as the parametric projection algorithm within Visual to Parametric Interaction systems.
Description
Keywords
Computer science, Contrastive Learning, Deep Learning, Human in the loop machine learning, Visual Analytics, Visual to Parametric Interaction
Citation
Extent
106 pages
Format
Geographic Location
Time Period
Related To
Table of Contents
Rights
All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.
Rights Holder
Local Contexts
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.