Modes of Gaussian Mixtures and an Inequality for the Distance Between Curves in Space
Date
2012
Authors
Advisors
Journal Title
Journal ISSN
Volume Title
Repository Usage Stats
views
downloads
Abstract
This dissertation studiess high dimensional problems from a low dimensional perspective. First, we explore rectifiable curves in high-dimensional space by using the Fréchet distance between and total curvatures of the two curves to bound the difference of their lengths. We create this bound by mapping the curves into R^2 while preserving the length between the curves and increasing neither
the total curvature of the curves nor the Fr\'echet distance between them. The bound is independent of the dimension of the ambient Euclidean space, it improves upon a bound by Cohen-Steiner and Edelsbrunner for dimensions greater than three and it generalizes
a result by F\'ary and Chakerian.
In the second half of the dissertation, we analyze Gaussian mixtures. In particular, we consider the sum of n Gaussians, where each Gaussian is centered at the vertex of a regular n-simplex. Fixing the width of the Guassians and varying the diameter of the simplex from zero to infinity by increasing a parameter that we call the scale factor, we find the window of scale factors for which the Gaussian mixture has more modes, or local maxima, than components of the mixture.
We see that the extra mode created is subtle, but can be higher than the modes closer to the vertices of the simplex. In addition, we prove that all critical points are located on a set of one-dimensional lines (axes) connecting barycenters of complementary faces of
the simplex.
Type
Department
Description
Provenance
Subjects
Citation
Permalink
Citation
Fasy, Brittany Terese (2012). Modes of Gaussian Mixtures and an Inequality for the Distance Between Curves in Space. Dissertation, Duke University. Retrieved from https://hdl.handle.net/10161/5793.
Collections
Except where otherwise noted, student scholarship that was shared on DukeSpace after 2009 is made available to the public under a Creative Commons Attribution / Non-commercial / No derivatives (CC-BY-NC-ND) license. All rights in student work shared on DukeSpace before 2009 remain with the author and/or their designee, whose permission may be required for reuse.