Date of Award

3-18-2013

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Applied Science

First Advisor

Keith Bush

Abstract

The purpose of this project is to determine if the properties preserved by low-dimensional manifolds of high-dimensional data sets can be combined with human visual acuity to improve search. Dimensionality-reducing manifold embeddings preserve relative path lengths between data elements even as a number of dimensions by which the data is represented are removed. By preserving smooth spatial variations between data elements, we may leverage the power of the human vision system to infer spatial patterns within the manifold and rapidly guide the search process to highly desirable regions of the data set: this should be true even when relevant data items are missing. Such a visual search tool would be useful for rapidly retrieving complex data items or for identifying similar but missing elements from a large database. Relevant applications could include information retrieval and forensics.

Download

Share

COinS