Computer Sciencehttp://hdl.handle.net/10012/99302018-07-19T17:37:02Z2018-07-19T17:37:02ZOn Skin Cyanotic Appearances and Spectral Responses Elicited by Methemoglobinemia and SulfhemoglobinemiaAskew, Stephenhttp://hdl.handle.net/10012/134782018-07-14T02:30:31Z2018-07-13T00:00:00ZOn Skin Cyanotic Appearances and Spectral Responses Elicited by Methemoglobinemia and Sulfhemoglobinemia
Askew, Stephen
Methemoglobinemia and sulfhemoglobinemia are potentially life-threatening blood disorders characterized by similar symptoms and markedly distinct treatment procedures. In this thesis, we investigate the causal relationship between these disorders and the onset of cyanosis (purple or bluish skin coloration). More specifically, we perform controlled experiments to elicit cyanotic appearances resulting from different severity levels of these disorders and varying physiological conditions. We note that such experiments cannot be induced in living subjects without posing risks to their health. Accordingly, we have resorted to an in silico experimental approach supported by biophysical data reported in the biomedical literature. Besides bringing new insights about cyanotic chromatic variations elicited by methemoglobinemia and sulfhemoglobinemia, our investigation provides the basis for the proposition of a cost-effective protocol for the noninvasive detection and differentiation of these disorders. Our experimental results indicate that its sensitivity range exceeds the range of similar technologies, which are in general associated with high operational costs. We believe that these aspects make the proposed protocol particularly suitable for incorporation into noninvasive disease screening/diagnostic systems, particularly those deployed at the point of care of medical settings with limited access to laboratory resources.
2018-07-13T00:00:00ZShortest Paths in Geometric Intersection GraphsSkrepetos, Dimitrioshttp://hdl.handle.net/10012/134542018-06-30T02:30:21Z2018-06-29T00:00:00ZShortest Paths in Geometric Intersection Graphs
Skrepetos, Dimitrios
This thesis studies shortest paths in geometric intersection graphs, which can model, among others, ad-hoc communication and transportation networks. First, we consider two classical problems in the field of algorithms, namely Single-Source Shortest Paths (SSSP) and All-Pairs Shortest Paths (APSP). In SSSP we want to compute the shortest paths from one vertex of a graph to all other vertices, while in APSP we aim to find the shortest path between every pair of vertices. Although there is a vast literature for these problems in many graph classes, the case of geometric intersection graphs has been only partially addressed.
In unweighted unit-disk graphs, we show that we can solve SSSP in linear time, after presorting the disk centers with respect to their coordinates. Furthermore, we give the first (slightly) subquadratic-time APSP algorithm by using our new SSSP result, bit tricks, and a shifted-grid-based decomposition technique.
In unweighted, undirected geometric intersection graphs, we present a simple and general technique that reduces APSP to static, offline intersection detection. Consequently, we give fast APSP algorithms for intersection graphs of arbitrary disks, axis-aligned line segments, arbitrary line segments, d-dimensional axis-aligned boxes, and d-dimensional axis-aligned unit hypercubes. We also provide a near-linear-time SSSP algorithm for intersection graphs of axis-aligned line segments by a reduction to dynamic orthogonal point location.
Then, we study two problems that have received considerable attention lately. The first is that of computing the diameter of a graph, i.e., the longest shortest-path distance between any two vertices. In the second, we want to preprocess a graph into a data structure, called distance oracle, such that the shortest path (or its length) between any two query vertices can be found quickly. Since these problems are often too costly to solve exactly, we study their approximate versions.
Following a long line of research, we employ Voronoi diagrams to compute a (1+epsilon)-approximation of the diameter of an undirected, non-negatively-weighted planar graph in time near linear in the input size and polynomial in 1/epsilon. The previously best solution had exponential dependency on the latter. Using similar techniques, we can also construct the first (1+epsilon)-approximate distance oracles with similar preprocessing time and space and only O(log(1/\epsilon)) query time.
In weighted unit-disk graphs, we present the first near-linear-time (1+epsilon)-approximation algorithm for the diameter and for other related problems, such as the radius and the bichromatic closest pair. To do so, we combine techniques from computational geometry and planar graphs, namely well-separated pair decompositions and shortest-path separators. We also show how to extend our approach to obtain O(1)-query-time (1+epsilon)-approximate distance oracles with near linear preprocessing time and space. Then, we apply these oracles, along with additional ideas, to build a data structure for the (1+epsilon)-approximate All-Pairs Bounded-Leg Shortest Paths (apBLSP) problem in truly subcubic time.
2018-06-29T00:00:00ZTowards Better Methods of Stereoscopic 3D Media Adjustment and StylizationIstead, Lesleyhttp://hdl.handle.net/10012/133642018-05-31T02:30:29Z2018-05-30T00:00:00ZTowards Better Methods of Stereoscopic 3D Media Adjustment and Stylization
Istead, Lesley
Stereoscopic 3D (S3D) media is pervasive in film, photography and art. However, working with
S3D media poses a number of interesting challenges arising from capture and editing. In this thesis
we address several of these challenges. In particular, we address disparity adjustment and present
a layer-based method that can reduce disparity without distorting the scene. Our method was
successfully used to repair several images for the 2014 documentary “Soldiers’ Stories” directed by
Jonathan Kitzen. We then explore consistent and comfortable methods for stylizing stereo images.
Our approach uses a modified version of the layer-based technique used for disparity adjustment
and can be used with a variety of stylization filters, including those in Adobe Photoshop. We
also present a disparity-aware painterly rendering algorithm. A user study concluded that our
layer-based stylization method produced S3D images that were more comfortable than previous
methods. Finally, we address S3D line drawing from S3D photographs. Line drawing is a common
art style that our layer-based method is not able to reproduce. To improve the depth perception of
our line drawings we optionally add stylized shading. An expert survey concluded that our results
were comfortable and reproduced a sense of depth.
2018-05-30T00:00:00ZAnalytics for EveryoneEl Gebaly, Kareemhttp://hdl.handle.net/10012/133502018-05-24T02:30:47Z2018-05-23T00:00:00ZAnalytics for Everyone
El Gebaly, Kareem
Analyzing relational data typically involves tasks that facilitate gaining familiarity or insights
and coming up with findings or conclusions based on the data. This process is usually practiced
by data experts, such as data scientists, who share their output with a potentially less expert
audience (everyone). Our goal is to enable everyone to participate in analyzing data rather than
passively consuming its outputs (analytics democratization). With today’s increasing availability
of data (data democratization) on the internet (web) combined with already widespread personal
computing capabilities such a goal is becoming more attainable. With the recent increase of
public data, i.e., Open Data, users without a technical background are keener than ever to analyze
new data sets that are relevant to wide sectors of society. An important example of Open Data is
the data released by governments all over the world, i.e., Open Government.
This dissertation focuses on two main challenges that would face data exploration scenarios
such as exploring open data found over the web. First, the infrastructure necessary for interactive
data exploration is costly and hard to manage, especially by users who do not have technical
knowledge. Second, the target users need guidance through the data exploration since there are
too many starting points.
To eliminate challenges related to managing infrastructure, we propose an in-browser SQL
engine (serverless), i.e., a portable database, which we call Afterburner. Afterburner achieves
comparable performance to native SQL engines given the same resources on modestly sized data
sets. Afterburner uses code generation techniques that target an optimization-amenable subset
of JavaScript and employs typed arrays for its columnar-based in-memory storage. In addition,
for databases that are too large for the browser, we propose a hybrid architecture to accelerate
the performance of data exploration tasks: a one-time SQL query that runs at the backend and
SQL queries running in the browser as per user’s interactions. Based on a simple hint by the
user, Afterburner automatically splits queries into two parts: a backend query that generates a
materialized view that is shipped to the browser, and a frontend query per subsequent interaction
occur locally against this view. Optimizing queries using local materialized views inside the
browser accelerates query latency without adding any complexity to the backend or the frontend.
One common theme among many data exploration tasks revolves around navigating the many
different ways to group the data, i.e., exploring the data cube. Thus, to guide the user through data
exploration, we apply an information-theoretic technique that picks the most informative parts
from the entire data cube of a relational table, which is called Explanation Tables. We evaluate the
efficiency and effectiveness of a sampling-based technique for generating explanation tables that
achieves comparable quality to an exhaustive technique that considers the entire data cube, with
a significant reduction in the run time. In addition, we introduce optimizations to explanation
tables to fit the modest resources available in the browser without any external dependencies.
In this, we present an SQL engine and a data exploration guidance tool that run entirely in
the browser. We view the techniques and the experiments presented here as a fully functional
and open-sourced proof of viability of our proposal. Our analytical stack is portable and works
entirely in the browser. We show that SQL and exploration guidance can be as accessible as a
web page, which opens the opportunity for more people to analyze data sets. Facilitating data
exploration for everyone is one step closer towards analytics democratization where everyone
can participate in data exploration, not just the experts.
2018-05-23T00:00:00Z