CS Theses & Dissertations 2017

For 2017 graduation dates (in alphabetical order by last name):

An intelligent multi-agent based detection framework for classification of android malware
Alam, Mohammed Shahidul
DOI : 10.14288/1.0340265
URI : http://hdl.handle.net/2429/59914
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Vuong

Smartphones play an important role in our day to day activities. Some of them include monitoring our health such as eating habits, sleep patterns and exercise schedule. The Android mobile operating system developed by Google is currently the most popular operating system for such smart devices. It is also the most vulnerable device due to its open nature of software installation, ability to dynamically load code during runtime, and lack of updates to known vulnerabilities even on popular versions of the system. Thus, securing such devices from malware that targets user privacy and monetary resources is paramount. In this thesis, we developed a context-aware multi-agent based framework targeted towards protecting Android devices. A malware detection technique has to be context-aware due to limited battery resources of mobile devices. In some cases however, battery utilization might become secondary. This includes scenarios where detection accuracy is given a higher priority over battery utilization. Thus, a detection framework has to be intelligent and flexible. To reach this goal, our framework relies on building multiple scalable context based models, and observing the behaviour patterns of Android devices by comparing to relevant pre-built models. We make use of machine learning classifiers that are more scalable to help classify features that could be used to detect malware by behaviour analysis. In this framework, the expensive analysis components utilizing machine learning algorithms are pushed to server side, while agents on the Android client are used mainly for context-aware feature gathering to transmit the information to server side classifiers for analysis, and to receive classification results from the server side agents.

Investigating software developers' understanding of open source software licensing
Almeida, DanielAraujo
DOI : 10.14288/1.0354559
URI : http://hdl.handle.net/2429/62756
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. G. Murphy

Feature Recommender : a large-scale in-situ study of proactive software feature recommendations
Ardekani, Kamyar
DOI : 10.14288/1.0339868
URI : http : //hdl.handle.net/2429/59761
Degree : Master of Science - MSc
Graduation Date : 2017-05
Supervisor : Dr. McGrenere

SAT modulo monotonic theories
Bayless, Samuel John Liam
DOI : 10.14288/1.0343418
URI : http://hdl.handle.net/2429/61062
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisors : Dr. Hu, Dr. Hoos

Satisfiability Modulo Theories (SMT) solvers are a class of efficient constraint solvers which form integral parts of many algorithms. Over the years, dozens of different Satisfiability Modulo Theories solvers have been developed, supporting dozens of different logics. However, there are still many important applications for which specialized SMT solvers have not yet been developed. We develop a framework for easily building efficient SMT solvers for previously unsupported logics. Our techniques apply to a wide class of logics which we call monotonic theories, which include many important elements of graph theory and automata theory. Using this SAT Modulo Monotonic Theories framework, we created a new SMT solver, MonoSAT. We demonstrate that MonoSAT improves the state of the art across a wide body of applications, ranging from circuit layout and data center management to protocol synthesis - and even to video game design.

Combating the cold-start user problem in collaborative filtering recommender systems
Biswas, Sampoorna
DOI : 10.14288/1.0340630
URI : http://hdl.handle.net/2429/60254
Degree : Master of Science – MSc
Graduation Date : 2017-05
Supervisor : Dr. Lakshmanan

Building believable robots: an exploration of how to make simple robots look, move, and feel right
Bucci, Paul Alexand Hendrik
DOI : 10.14288/1.0355204
URI : http://hdl.handle.net/2429/62866
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Maclean

Cross-platform data integrity and confidentiality with graduated access control
Chen, Feifan
DOI: 10.14288/1.0340667
UR I: http://hdl.handle.net/2429/60265
Degree : Master of Science - MSc
Graduation Date : 2017-05
Supervisor : Dr. Beschastnikh

Embodied Perception during Walking using Deep Recurrent Neural Networks
Chen, Jacob
DOI : 10.14288/1.0348730
URI : http://hdl.handle.net/2429/62171
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. van de Panne

Deep kernel mean embeddings for generative modeling and feedforward style transfer
Chen, Tian Qi
DOI : 10.14288/1.0354397
URI : http://hdl.handle.net/2429/62668
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Schmidt

Extending systems with virtual hardware aggregation
Cully, Brendan
DOI : 10.14288/1.0342739
URI : http : //hdl.handle.net/2429/60579
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Warfield

Hardware has physical limitations. It has fixed performance limits, and may fail. Applications suffer from the limitations of the physical hardware on which they run. Making applications able to take advantage of multiple hardware instances to avoid these limitations is complex. Since this effort must be expended for every application, it is impractical for most of them. In this thesis, we show that we can aggregate multiple physical machines at the virtual machine interface, allowing them to transcend the limitations of single machines without changing the applications themselves.

Topological Modeling for Vector Graphics
Dalstein, Boris
DOI : 10.14288/1.0355734
URI : http://hdl.handle.net/2429/63078
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. van de Panne

In recent years, with the development of mobile phones, tablets, and web technologies, we have seen an ever-increasing need to generate vector graphics content, that is, resolution-independent images that support sharp rendering across all devices, as well as interactivity and animation. However, the tools and standards currently available to artists for authoring and distributing such vector graphics content have many limitations. Importantly, basic topological modeling, such as the ability to have several faces share a common edge, is largely absent from current vector graphics technologies. In this thesis, we address this issue with three major contributions. First, we develop theoretical foundations of vector graphics topology, grounded in algebraic topology. More specifically, we introduce the concept of Point-Curve-Surface complex (PCS complex) as a formal tool that allows us to interpret vector graphics illustrations as non-manifold, non-planar, non-orientable topological spaces immersed in R2, unlike planar maps which can only represent embeddings. Second, based on this theoretical understanding, we introduce the vector graphics complex (VGC) as a simple data structure that supports fundamental topological modeling operations for vector graphics illustrations. It allows for the direct representation of incidence relationships between objects, while at the same time keeping the geometric flexibility of stacking-based systems, such as the ability to have edges and faces overlap each others. Third and last, based on the VGC, we introduce the vector animation complex (VAC), a data structure for vector graphics animation, designed to support the modeling of time-continuous topological events, which are common in 2D hand-drawn animation. This allows features of a connected drawing to merge, split, appear, or disappear at desired times via keyframes that introduce the desired topological change. Because the resulting space-time complex directly captures the time-varying topological structure, features are readily edited in both space and time in a way that reflects the intent of the drawing.

Computational projection display - towards efficient high brightness projection in cinema
Damberg, Gerwin
DOI : 10.14288/1.0349061
URI : http://hdl.handle.net/2429/62421
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. Heidrich

Cinema projectors need to compete with home theater displays in terms of image quality. High frame rate and high spatial resolution as well as stereoscopic 3D are common features today, but even the most advanced cinema projectors lack in-scene contrast and more importantly high peak luminance, both of which are essential perceptual attributes for images to look realistic. At the same time studies on HDR image statistics suggest that the average image intensity in a controlled ambient viewing environment such as cinema can be as low as 1% for cinematic HDR content and does not often exceed 18%, middle gray in photography. Traditional projection systems form images and colours by blocking the source light from a lamp, therefore attenuating on average between 99% and 82% of light before it reaches the screen. This inefficient use of light poses significant challenges for achieving higher peak brightness levels. We propose a new projector architecture built around commercially available components, in which light can be steered to form images. The gain in system efficiency significantly reduces the total cost of ownership of a projector (fewer components and lower operating cost) and at the same time increases peak luminance and improves black level beyond what is practically achievable with incumbent projector technologies. At the heart of this computational display technology is a new projector hardware design using phase-modulation in combination with new optimization algorithms for real-time phase retrieval. Based on this concept we propose and design a full featured projector prototype. To allow for display of legacy SDR as well as high brightness HDR content on light steering projectors we derive perceptually motivated, calibrated tone mapping and colour appearance models. We develop a calibrated optical forward model of the projector hardware and analyse the impact of content mapping parameters and algorithm choices on (light) power requirements.

Estimating cell type proportions in human cord blood samples from DNAm arrays
Dinh, Louie
DOI : 10.14288/1.0356611
URI : http://hdl.handle.net/2429/63224
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisors : Dr. Ng, Dr. Mostafavi (Departments of Statistics & Medical Genetics)

[no title]
Dong, Zheng
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Lakshmanan

[no title]
Dong, Wenqiang
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Munzner

Finding a record in a database
Fatemi, Bahare
DOI : 10.14288/1.0353194
URI : http://hdl.handle.net/2429/62575
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Poole

FlowRep: extracting descriptive curve networks from free-form design shapes
Gori, Giorgio
DOI : 10.14288/1.0353171
URI : http://hdl.handle.net/2429/62558
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Sheffer

Study of two biochemical models: chemical reaction networks, and nucleic acid systems
Hajiaghayi, Monir
DOI : 10.14288/1.0357130
URI : http://hdl.handle.net/2429/63311
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. Condon

The contributions of this thesis are motivated by an exciting challenge at the intersection of computer science and biochemistry: Can we program molecules to do interesting or useful computations? There has been significant progress in programming nucleic acids - particularly DNA molecules - thanks in part to availability of models and algorithms for predicting nucleic acid structure and folding kinetics. At a higher level of abstraction, Chemical Reaction Networks (CRNs) have proven to be valuable as a molecular programming model that enables researchers to understand the potential and limitations of computing with molecules, unencumbered by low-level details. These two levels of abstraction are linked; it is possible to "compile" CRN programs into nucleic acid systems that make the programs implementable in a test tube. We design and analyze CRN algorithms for two purposes. First, we show how any semilinear function can be computed by CRNs, even when no "leader" species (i.e., initial species with constant but non-zero counts) is present. Our results improve earlier results of Chen et al. (2012) who showed that only semilinear functions are computable by error-free CRNs using leaders. Our new CRN construction can be done in expected time O(n), which is faster than O(n log n) bound achieved by Chen et al. Second, we provide the most intuitive proofs of correctness and efficiency for three different CRNs computing Approximate Majority: Given a mixture of two types of species with an initial gap between their counts, a CRN computation must reach totality on the majority species with high probability. The CRNs of our interest have the ability to start with an initial gap of Ω(√n log n). In the second part of this thesis, we study the problem of predicting the Minimum Free Energy secondary structure (the set of base pairs) of a given set of nucleic acid strands with no pseudoknots (crossing base pairs). We show that this problem is APX-hard which implies that there does not exist a polynomial time approximation scheme for this problem, unless P = NP. We also propose a new Monte-Carlo based method to efficiently estimate nucleic acid folding kinetics.

Validation of SQL queries over streaming warehouses
Jain, Ritika
DOI : 10.14288/1.0355210
URI : http://hdl.handle.net/2429/62867
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Lakshmanan

Improve classification on infrequent discourse relations via training data enrichment
Jiang, Kailang
DOI : 10.14288/1.0340024
URI : http : //hdl.handle.net/2429/59844
Degree : Master of Science - MSc
Graduation Date : 2017-05
Supervisors : Dr. Carenini, Dr. Ng

Data Mining Approach for Adding Adaptive Interventions to Exploratory Learning Environments
Kardan, Samad
DOI : 10.14288/1.0348694
URI : http://hdl.handle.net/2429/62133
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. Conati

Due to the open-ended nature of the interaction with Exploratory Learning Environments (ELEs), it is not trivial to add mechanisms for providing adaptive support to users. Our goal is to devise and evaluate a data mining approach for providing adaptive interventions that help users to achieve better task performance during interaction with ELEs. The general idea of this thesis is as follows: In an exploratory and open-ended environment, we collect interaction data of users while they are working with the system, and then find representative patterns of behavior for different user groups that achieved various levels of task performance. We use these patterns to provide adaptive real-time interventions designed to suggest or enforce the effective interaction behaviors while discouraging or preventing the ineffective ones. We test and confirm the hypothesis that as a result of these interventions, the average learning performance of the new users who work with the adaptive version of this ELE is significantly higher than the non-adaptive version. We use an interactive simulation for learning Constraint Satisfaction Problems (CSP), the AIspace CSP applet, as the test-bed for our research and propose a framework which covers the entire process described above, called the User Modeling and Adaptation (UMA) framework. The contributions of this thesis are two-fold: i) It contributes to the Educational Data Mining (EDM) research, by devising, modifying, and testing different techniques and mechanisms for a complete data mining based approach to delivering adaptive interventions in ELEs summarized in the UMA framework. The UMA framework consists of 3 phases: Behavior Discovery, User Classification, and Adaptive Support. We assessed each of the above phases in a series of user studies. This work is the first to fully evaluate and provide positive evidence for the use of a data mining approach for deriving and delivering adaptive interventions in ELEs with the goal of improving the user’s performance. ii) It also contributes to the user modeling and user-adapted interaction community by providing new evidence for the usefulness of the eye-gaze data for the purpose of predicting learning performance of users while interacting with an ELE.

Surface based fluid animation using integral equations: simulation and compression
Keeler, Todd Douglas
DOI : 10.14288/1.0355876
URI : http://hdl.handle.net/2429/63183
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. Bridson

This dissertation looks at exploiting the mathematics of vorticity dynamics and potential flow using integral equations to reformulate critical parts of fully dynamic fluid animation methods into surface based problems. These reformulations enable more efficient calculation and data-structures due to the reduction of the simulation domain to the two dimensional fluid surface, rather than its volume. We also introduce a surface compression and real-time playback method for continuous time-dependent iso-surfaces. This compression method further increases the impact of our highly efficient surface-based simulation methods.

Mining unstructured social streams: cohesion, context and evolution
Li, Pei
DOI : 10.14288/1.0343307
URI : http://hdl.handle.net/2429/60982
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Lakshmanan

As social websites like Twitter greatly influence people's digital life, unstructured social streams become prevalent, which are fast surging textual post streams without formal structure or schema between posts or inside the post content. Modeling and mining unstructured social streams in Twitter become a challenging and fundamental problem in social web analysis, which leads to numerous applications, e.g., recommending social feeds like "what's happening right now?" or "what are related stories?". Current social stream analysis in response to queries merely return an overwhelming list of posts, with little aggregation or semantics. The design of the next generation social stream mining algorithms faces various challenges, especially, the effective organization of meaningful information from noisy, unstructured, and streaming social content. The goal of this dissertation is to address the most critical challenges in social stream mining using graph-based techniques. We model a social stream as a post network, and use "event" and "story" to capture a group of aggregated social posts presenting similar content in different granularities, where an event may contain a series of stories. We highlight our contributions on social stream mining from a structural perspective as follows. We first model a story as a quasi-clique, which is cohesion-persistent regardless of the story size, and propose two solutions, DIM and SUM, to search the largest story containing given query posts, by deterministic and stochastic means, respectively. To detect all stories in the time window of a social stream and support the context-aware story-telling, we propose CAST, which defines a story as a (k,d)-Core in post network and tracks the relatedness between stories. We propose Incremental Cluster Evolution Tracking (ICET), which is an incremental computation framework for event evolution on evolving post networks, with the ability to track evolution patterns of social events as time rolls on. Approaches in this dissertation are based on two hypotheses: users prefer correlated posts to individual posts in post stream modeling, and a structural approach is better than frequency/LDA-based approaches in event and story modeling. We verify these hypotheses by crowdsourcing based user studies.

Characterizing minimum-length coordinated motions of two disks
Liu, Paul
DOI : 10.14288/1.0345619
URI : http://hdl.handle.net/2429/61351
Degree : Master of Science – MSc
Graduation Date : 2017-05
Supervisor : Dr. Kirkpatrick

Vaportrail: a platform for personal data applications
MacRow, Kalan William
DOI : 10.14288/1.0354256
URI : http://hdl.handle.net/2429/62599
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Warfield

Distributed linear programming with Apache Spark
Mohyedin Kermani, Ehsan
DOI : 10.14288/1.0340337
URI : http://hdl.handle.net/2429/59990
Degree : Master of Science - MSc
Graduation Date : 2017-05
Supervisors : Dr. Greif, Dr. Ascher

A semi-joint neural model for sentence level discourse parsing and sentiment analysis
Nejat, Bita
DOI : 10.14288/1.0354697
URI : http://hdl.handle.net/2429/62824
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Carenini, Dr. Ng

Efficient feasibility checking in reverse clock auctions for radio spectrum
Newman, Neil
DOI : 10.14288/1.0340850
URI : http : //hdl.handle.net/2429/60376
Degree : Master of Science - MSc
Graduation Date : 2017-05
Supervisor : Dr. Leyton-Brown

Developing Locomotion Skills with Deep Reinforcement Learning
Peng, Xue Bin
DOI : 10.14288/1.0345638
URI : http://hdl.handle.net/2429/61370
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. van de Panne

Visual text analytics for online conversations
Prince, Md Enamul Hoque
DOI : 10.14288/1.0347620
URI : http://hdl.handle.net/2429/61772
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. Carenini

With the proliferation of Web-based social media, asynchronous conversations have become very common for supporting online communication and collaboration. Yet the increasing volume and complexity of conversational data often make it very difficult to get insights about the discussions. This dissertation posits that by integrating natural language processing and information visualization techniques in a synergistic way, we can better support the user's task of exploring and analyzing conversations. Unlike most previous systems, which do not consider the specific characteristics of online conversations; we applied design study methodologies from the visualization literature to uncover the data and task abstractions that guided the development of a novel set of visual text analytics systems. The first of such systems is ConVis, that supports users in exploring an asynchronous conversation, such as a blog. ConVis offers a visual overview of a conversation by presenting topics, authors, and the thread structure of a conversation, as well as various interaction techniques such as brushing and linked highlighting. Broadening from a single conversation to a collection of conversations, MultiConVis combines a novel hierarchical topic modeling with multi-scale exploration techniques. A series of user studies revealed the significant improvements in user performance and subjective measures when these two systems were compared to traditional blog interfaces. Based on the lessons learned from these studies, this dissertation introduced an interactive topic modeling framework specifically for asynchronous conversations. The resulting systems empower the user in revising the underlying topic models through an intuitive set of interactive features when the current models are noisy and/or insufficient to support their information seeking tasks. Two summative studies suggested that these systems outperformed their counterparts that do not support interactive topic modeling along several subjective and objective measures. Finally, to demonstrate the generality and applicability of our approach, we tailored our previous systems to support information seeking in community question answering forums. The prototype was evaluated through a large-scale Web-based study, which suggests that our approach can be adapted to a specific conversational genre among a diverse range of users. The dissertation concludes with a critical reflection on our approach and considerations for future research.

Toward user-adaptive visualizations: further results on real-time prediction of user cognitive abilities from action and eye-tracking data
Rahman, Md Abed
DOI : 10.14288/1.0348690
URI : http://hdl.handle.net/2429/62128
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Conati

Perception of motion in virtual reality interception tasks
Rolin, Robert Adam
DOI : 10.14288/1.0354558
URI : http://hdl.handle.net/2429/62755
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Pai

Haptic experience design : tools, techniques, and process
Schneider, Oliver Stirling
DOI : 10.14288/1.0340617
URI : http : //hdl.handle.net/2429/60233
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Maclean

Haptic technology, which engages the sense of touch, offers promising benefits for a variety of interactions including low-attention displays, emotionally-aware interfaces, and augmented media experiences. Despite an increasing presence of physical devices in commercial and research applications, there is still little support for the design of engaging haptic sensations. Previous literature has focused on the significant challenges of technological capabilities or physical realism rather than on supporting experience design. In this dissertation, we study how to design, build, and evaluate interactive software to support haptic experience design (HaXD). We define HaXD and iteratively design three vibrotactile effect authoring tools, each a case study covering a different user population, vibrotactile device, and design challenge, and use them to observe specific aspects of HaXD with their target users. We make these in-depth findings more robust in two ways :  generalizing results to a breadth of use cases with focused design projects, and grounding them with expert haptic designers through interviews and a workshop. Our findings 1) describe HaXD, including processes, strategies, and challenges; and 2) present guidelines on designing, building, and evaluating interactive software that facillitates HaXD. When characterizing HaXD processes, strategies, and challenges, we show that experience design is already practiced with haptic technology, but faces unique considerations compared to other modalities. We identify four design activities that must be explicitly supported :  sketching, refining, browsing, and sharing. We find and develop strategies to accommodate the wide variety of haptic devices. We articulate approaches for designing meaning with haptic experiences, and finally, highlight a need for supporting adaptable interfaces. When informing the design, implementation, and evaluation of HaXD tools, we discover critical features, including a need for improved online deployment and community support. We present steps to develop both existing and future research software into a mature suite of HaXD tools, and reflect upon evaluation methods. By characterizing HaXD and informing supportive tools, we make a first step towards establishing HaXD as its own field, akin to graphic and sound design.

Personalizing haptics: from individuals’ sense-making schemas to end-user haptic tools
Seifi, Hasti
DOI : 10.14288/1.0343521
URI : http://hdl.handle.net/2429/61165
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Maclean

Synthetic haptic sensations will soon proliferate throughout many aspects of our lives, well beyond the simple buzz we get from our mobile devices. This view is widely held, as evidenced by the growing list of use cases and industry's increasing investment in haptics. However, we argue that taking haptics to the crowds will require haptic design practices to go beyond a one-size-fits-all approach, common in the field, to satisfy users' diverse perceptual, functional, and hedonic needs and preferences reported in the literature. In this thesis, we tackle end-user personalization to leverage utility and aesthetics of haptic signals for individuals. Specifically, we develop effective haptic personalization mechanisms, grounded in our synthesis of users' sense-making schemas for haptics. First, we propose a design space and three distinct mechanisms for personalization tools: choosing, tuning, and chaining. Then, we develop the first two mechanisms into: 1) an efficient interface for choosing from a large vibration library, and 2) three emotion controls for tuning vibrations. In developing these, we devise five haptic facets that capture users' cognitive schemas for haptic stimuli, and derive their semantic dimensions and between-facet linkages by collecting and analyzing users' annotations for a 120-item vibration library. Our studies verify utility of the facets as a theoretical model for personalization tools. In collecting users' perception, we note a lack of scalable haptic evaluation methodologies and develop two methodologies for large-scale in-lab evaluation and online crowdsourcing of haptics. Our studies focus on vibrotactile sensations as the most mature and accessible haptic technology but our contributions extend beyond vibrations and inform other categories of haptics.

Investigating completeness and consistency of links between issues and commits
Thompson, Casey Albert
DOI : 10.14288/1.0354567
URI : http://hdl.handle.net/2429/62775
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. G. Murphy

Classification of puck possession events in ice hockey
Tora, Moumita Roy
DOI : 10.14288/1.0355849
URI : http://hdl.handle.net/2429/63149
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Little

Improved action and path synthesis using gradient sampling
Traft, Neil Michael
DOI : 10.14288/1.0357150
URI : http://hdl.handle.net/2429/63328
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Mitchell

Towards large-scale nonparametric scene parsing of images and video
Tung, Frederick
DOI : 10.14288/1.0343064
URI : http://hdl.handle.net/2429/60790
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Little

In computer vision, scene parsing is the problem of labelling every pixel in an image or video with its semantic category. Its goal is a complete and consistent semantic interpretation of the structure of the real world scene. Scene parsing forms a core component in many emerging technologies such as self-driving vehicles and prosthetic vision, and also informs complementary computer vision tasks such as depth estimation. This thesis presents a novel nonparametric scene parsing framework for images and video. In contrast to conventional practice, our scene parsing framework is built on nonparametric search-based label transfer instead of discriminative classification. We formulate exemplar-based scene parsing for both 2D (from images) and 3D (from video), and demonstrate accurate labelling on standard benchmarks. Since our framework is nonparametric, it is easily extensible to new categories and examples as the database grows. Nonparametric scene parsing is computationally demanding at test time, and requires methods for searching large collections of data that are time and memory efficient. This thesis also presents two novel binary encoding algorithms for large-scale approximate nearest neighbor search: the bank of random rotations is data independent and does not require training, while the supervised sparse projections algorithm targets efficient search of high-dimensional labelled data. We evaluate these algorithms on standard retrieval benchmarks, and then demonstrate their integration into our nonparametric scene parsing framework. Using 256-bit codes, binary encoding reduces search times by an order of magnitude and memory requirements by three orders of magnitude, while maintaining a mean per-class accuracy within 1% on the 3D scene parsing task.

"Not able to resist the urge" : social insider attacks on Facebook
Usmani, Wali Ahmed
DOI : 10.14288/1.0340682
URI : http : //hdl.handle.net/2429/60294
Degree : Master of Science - MSc
Graduation Date : 2017-05
Supervisors : Dr. Beschastnikh, Dr. Beznosov (Electrical & Computer Engineering)

Designing zooming interactions for small displays with a proximity sensor
Ustek, Dilan
DOI : 10.14288/1.0354394
URI : http://hdl.handle.net/2429/62656
Degree : Master of Science – MSc
Graduation Date : 2017-11
Supervisor : Dr. Maclean

Storage system design for fast nonvolatile memories
Wires, Jake Taylor
DOI : 10.14288/1.0357126
URI : http://hdl.handle.net/2429/63305
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisor : Dr. Warfield

Nonvolatile memories are transforming the data center. Over the past decade, enterprise flash has evolved to provide a thousand times more random-access throughput than mechanical disks, with a thousand times lower latency and ten times more capacity. These remarkable improvements completely reshape software concerns, allowing storage systems to take a more central role in dynamic resource management, but demanding that they do so with significantly lower overheads. This thesis presents several novel software techniques for managing high-density storage systems. In particular, it describes a probabilistic approach to workload modeling that provides guaranteed error bounds while dramatically reducing memory overheads relative to existing state-of-the-art algorithms. It also documents the design and implementation of a storage controller that leverages dynamic constraint satisfaction techniques to continually optimize data and network flow placement for performance, efficiency, and scale. These advances are presented within a broader design framework that provides a flexible and robust platform for managing all aspects of storage resource allocation. Informed by experiences and insights gained over six years of building an enterprise scale-out storage appliance, it is based on three key ideas: light-weight abstraction to decouple logical resources from physical hardware, online analysis to capture workload requirements, and dynamic actuation to adjust allocations as requirements change. Together, these capabilities allow storage software to dynamically adapt to changing workload behavior and allow stored data to play a more active role in data center computing.

Methods for image recovery in computational imaging
Xiao, Lei
DOI : 10.14288/1.0343606
URI : http://hdl.handle.net/2429/61230
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-05
Supervisor : Dr. Heidrich

As a classic topic that has been studied for decades, image restoration is still a very active research area. Developing more effective and efficient methods are highly desirable. This thesis addresses image restoration problems for applications in computational imaging including Time-of-Flight (ToF) imaging and digital photography. While ToF cameras have shown great promise at low-cost depth imaging, they suffer from limited depth-of-field and low spatial-resolution. We develop a computational method to remove lens blur and increase image resolution of off-the-shelf ToF cameras. The method solves latent images directly from the raw sensor data as an inverse problem, and supports for future ToF cameras that use multiple frequencies, phases and exposures. Photographs taken by hand-held cameras are likely to suffer from blur caused by camera shake during exposure. Removing such blur and recovering sharp images as a post-process is therefore critical. We develop a blind deblurring method that is purely based on stochastic random-walk optimization. This simple framework in combination with different priors produces comparable results to the much more complex state-of-the-art deblurring algorithms. Blur causes even more serious issues for document photographs as slight blur can make Optical Character Recognition (OCR) techniques fail. We address the blind deblurring problem specifically for common document photographs. Observing that the latter are mostly composed of high-order structures, our method captures such domain property by a series of high-order filters as well as customized response functions. These parameters are trained from data by discriminative learning approach and form an end-to-end network that can efficiently and jointly estimate blur kernels and legible images. Discriminative learning approaches achieve convincing trade-off between image quality and computational efficiency, however, they require separate training for each restoration task and problem condition, making it time-consuming and difficult to encompass all tasks and conditions during training. We combine discriminative learning and formal optimization techniques to learn image priors that require a single-pass training and share across various tasks and conditions while keeping the efficiency as previous discriminative methods. After being trained, our method can be combined with other likelihood or priors to address unseen restoration tasks or further improve the image quality.

Tackling small-scale fluid features in large domain for computer graphics
Zhang, Xinxin
DOI : 10.14288/1.0347254
URI : http://hdl.handle.net/2429/61435
Degree : Doctor of Philosophy - PhD
Graduation Date : 2017-11
Supervisors : Dr. Bridson, Dr. Greif

Turbulent gaseous phenomena, often visually characterized by their swirling nature, are mostly dominated by the evolution of vorticity. Small scale vortex structures are essential to the look of smoke, fire, and related effects, whether produced by vortex interactions, jumps in density across interfaces (baroclinity), or viscous boundary layers. Classic Eulerian fluid solvers do not cost-effectively capture these small-scale features in large domains. Lagrangian vortex methods show great promise from turbulence modelling, but face significant challenges in handling boundary conditions, making them less attractive for computer graphics applications. This thesis proposes several novel solutions for the efficient simulation of small scale vortex features both in Eulerian and Lagrangian frameworks, extending robust Eulerian simulations with new algorithms inspired by Lagrangian vortex methods.