Please note: This PhD defence will be given online.
Bryce
Sandlund, PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Supervisor: Professor J. Ian Munro
This thesis considers the study of data structures from the perspective of the theoretician, with a focus on simplicity and practicality. We consider both the time complexity as well as space usage of proposed solutions. Topics discussed fall in three main categories: partial order representation, range modes, and graph cuts.
We consider two problems in partial order representation. The first is a data structure to represent a lattice. A lattice is a partial order where the set of elements larger than any two elements x and y are all larger than an element z, known as the join of x and y; a similar condition holds for elements smaller than any two elements. Our data structure is the first correct solution that can simultaneously compute joins and the inverse meet operation in sublinear time while also using subquadratic space. The second is a data structure to support queries on a dynamic set of one-dimensional ordered data; that is, essentially any operation computable on a binary search tree. We develop a data structure that is able to interpolate between binary search trees and efficient priority queues, offering more-efficient insertion times than the former when query distribution is non-uniform.
We also consider static and dynamic exact and approximate range mode. Given one-dimensional data, the range mode problem is to compute the mode of a subinterval of the data. In the dynamic range mode problem, insertions and deletions are permitted. For the approximate problem, the element returned is to have frequency no less than a factor (1+epsilon) of the true mode, for some epsilon > 0. Our results include a linear-space dynamic exact range mode data structure that simultaneously improves on best previous operation complexity and an exact dynamic range mode data structure that breaks the Theta(n^(2/3)) time per operation barrier. For approximate range mode, we develop a static succinct data structure offering a logarithmic-factor space improvement and give the first dynamic approximate range mode data structure. We also consider approximate range selection.
The final category discussed is graph and dynamic graph algorithms. We develop an optimal offline data structure for dynamic 2- and 3- edge and vertex connectivity. Here, the data structure is given the entire sequence of operations in advance, and the dynamic operations are edge insertion and removal. Finally, we give a simplification of Karger's near-linear time minimum cut algorithm, utilizing heavy-light decomposition and iteration in place of dynamic programming in the subroutine to find a minimum cut of a graph G that cuts at most two edges of a spanning tree T of G.
To join this PhD defence on Zoom, please go to https://us02web.zoom.us/j/88425340145?pwd=OGVVQmZ1MERONW5TS2s2aW1jVjdOUT09.