CS Theses & Dissertations 2022

For 2022 graduation dates (in alphabetical order by last name):

Bootstrapping human optical flow and pose
Arko, Aritro Roy
DOI : 10.14288/1.0417462
URI : http://hdl.handle.net/2429/82442
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Kwang Moo Yi & Dr. Jim Little

Welfare maximization for complementary and competing items and their applications: analysis and algorithm using a utility driven model for achieving better social influence
Banerjee, Prithu
DOI : 10.14288/1.0412622
URI : http://hdl.handle.net/2429/81107
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-05
Supervisor : Dr. Laks Lakshmanan

Motivated by applications such as viral marketing, the problem of influence maximization (IM) has been extensively studied in the literature. The goal is to select a small number of users to adopt an item such that it results in a large cascade of adoptions by others. Existing works have three key limitations. (1) They do not account for the economic considerations of a user in buying/adopting items. (2) They cannot model the complex interactions between multiple items. (3) For the network owner, maximizing social welfare is important to ensure customer loyalty, which is not addressed in prior work in the IM literature. In this work, we address all three limitations and propose a novel model called Utility driven Independent Cascade (UIC) that combines utility-driven item adoption with influence propagation over networks. We focus on several types of items such as mutually complementary only, competing only, and a mix of the two in the context of the filter bubble problem. We formulate the problem of social welfare maximization under each of these settings. We show that while the objective function is neither submodular nor supermodular, a constant or instance-dependent approximation can still be achieved. With comprehensive experiments on real and synthetic datasets, we demonstrate that our algorithms significantly outperform all baselines on large real social networks.

An experimental study of the incomplete Cholesky factorization for the anisotropic diffusion equation 
Beauchamp, Austin Gregory
DOI : 10.14288/1.0418594
URI : http://hdl.handle.net/2429/82657
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Chen Greif

Efficient Computations on Uncertain Graphs by Group Testing, Streaming and Recycling 
Bevilacqua, Glenn Steven
DOI : 10.14288/1.0412921
URI : http://hdl.handle.net/2429/81252
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-05
Supervisor : Dr. Laks Lakshmanan

Uncertain graphs, where the presence of connections between nodes is probabilistic, have received a great deal of attention in a wide range of fields. Despite the progress made by prior works, the application of existing algorithms to solve the important problems of coverage maximization, also known as Influence Maximization (IM), and reachability estimation, also known as reliability, on uncertain graphs remains constrained by their computational costs in the form of both the running time and memory needed for one to achieve high quality solutions. In Chapter 2 we address the issue that when performing sampling on large networks the majority of random draws of edges are being effectively wasted as the majority of edges will not be live. We resolve this by introducing an approach that through the application of group testing of edges scales only logarithmically with the number of failed edges in the worst case as opposed to linearly. In Chapter 3 we tackle the problem of the exorbitant memory required to store the Reverse Influence Sample (RIS) collection used by existing approaches to solve the IM problem becoming prohibitive. We avoid this by developing a new non-greedy approach that avoids storing the RIS collection by streaming it instead. In Chapter 4 we note that a key cause of Monte Carlo simulation, along with existing approaches that build upon it, becoming ineffective for estimating low probability reachability is caused by the number of samples it needs to attain a desired relative error depending on the probability that is being estimated. We remedy this by developing a technique of recycling of random draws which enables us to develop an algorithm whose relative error does not depend on the probability being estimated and as such does not suffer from this limitation. In all cases we perform experiments on real datasets to empirically validate the effectiveness of the algorithms and techniques we develop.

Writers want AI collaborators to respect their personal values and writing strategies: a human-centered perspective on AI co-writing
Biermann, Oloff Carl
DOI : 10.14288/1.0420422
URI : http://hdl.handle.net/2429/82746
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Dongwook Yoon

Analysis and Preconditioning of Double Saddle-Point Systems
Bradley, Susanne Michelle
DOI : 10.14288/1.0417284
URI : http://hdl.handle.net/2429/82337
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-11
Supervisor : Dr. Chen Greif

This thesis deals with the mathematical analysis and numerical solution of double saddle-point systems. We derive bounds on the eigenvalues of a generic form of double saddle-point matrices with a positive definite leading block. The bounds are expressed in terms of extremal eigenvalues and singular values of the associated block matrices. Inertia and algebraic multiplicity of eigenvalues are considered as well. The analysis includes bounds for preconditioned matrices based on block diagonal preconditioners using Schur complements, and it is shown that in this case the eigenvalues are clustered within a few intervals bounded away from zero. Analysis for approximations of Schur complements is included. Some numerical observations validate our analytical findings. We also derive bounds on the eigenvalues of (classical) saddle-point matrices with singular leading blocks. The technique of proof is based on augmentation. Our bounds depend on the principal angles between the ranges or kernels of the matrix blocks. We use these analyses to derive a preconditioner for saddle-point systems with singular leading blocks. Our preconditioning approach is based on augmenting the leading block and using Schur complements of the augmented system. We show that the resulting preconditioned operator has four distinct eigenvalues, and numerical experiments validate the effectiveness of our approach. We then extend the preconditioner for saddle-point systems with a singular leading block to deal with double saddle-point systems with a singular leading block. The preconditioner is based on augmenting the leading block by a null matrix of one of the off-diagonal blocks, and using the Schur complements of the augmented system.

Sized dependent types via extensional type theory
Chan, Jonathan Ho-Wing
DOI : 10.14288/1.0416401
URI : http://hdl.handle.net/2429/82209
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. William Bowman

Supporting a developer's discovery of task-relevant information 
De Sousa Marques, Arthur
DOI : 10.14288/1.0421314
URI : http://hdl.handle.net/2429/82899
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-11
Supervisor : Dr. Gail Murphy

The information that a developer seeks to aid in the completion of a task typically exists across different kinds of software artifacts that include substantial natural language text. For instance, artifacts vary from conversational discussions about bug reports to tutorial descriptions of features in a library. In the artifacts that a developer consults, only some portions of the text will be useful to a developer's task and locating such portions can be time-consuming as the artifacts can include substantial text to peruse organized in different ways. For example, it might be easier to locate information in tutorial artifacts with structured headings whereas artifacts consisting of developer conversations might need to be read in detail. Given the limited time developers have to spend on any task, researchers have attempted to aid the developers by proposing a range of techniques to automate the identification of relevant text. However, this prior work is generally constrained to one or only a few types of artifacts. Enabling a developer access to artifact-specific approaches is difficult to deploy and support to the multitude of artifact types that is constantly evolving is challenging, if not impractical. In this dissertation, we propose a set of generalizable techniques to aid developers in locating a portion of text that might be useful for a task. These techniques are based on semantic patterns that arise from the empirical analysis of the text relevant to a task in multiple kinds of artifacts, leading us to propose techniques that incorporate the semantics of words and sentences to identify text likely relevant to a developer's task automatically. We evaluate the proposed techniques assessing the extent to which they identify text that developers deem relevant in different kinds of artifacts associated with Android development tasks. We then investigate how a tool that embeds the most promising semantic-based technique might assist developers while they perform a task. Results show that semantic-based techniques perform equivalently well across multiple artifact types and that a tool that automates the provision of task-relevant text assists developers in effectively completing a software development task.

myWeekInSight: visualizing personal data for chronic pain management through youth-centered design             
Desai, Unma Mayur
DOI : 10.14288/1.0419384
URI : http://hdl.handle.net/2429/82699
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Karon Maclean & Dr. Tim Oberlander (Health in Populations)

What does it take to be a haptician? : how community can empower designers and expose the many ways of being an expert 
Elbaggari, Hannah
DOI : 10.14288/1.0417559
URI : http://hdl.handle.net/2429/82538
Degree : Master of Science - MSc
Graduation Date : 2022-11
Supervisor : Dr. Karon Maclean

Duality in structured and federated optimization : theory and applications 
Fan, Zhenan
DOI : 10.14288/1.0421272
URI : http://hdl.handle.net/2429/82859
Degree : Doctor of Philosophy – PhD
Graduation Date : 2022-11
Supervisor : Dr. Michael Friedlander

The dual approach in mathematical optimization refers to a class of techniques for tackling a dual problem that arises from the original problem. Numerous notable improvements in strengthening the dual approach have been promoted in the last two decades because of its superior performance for many large-scale optimization problems. In this thesis, we investigate and extend the dual approach to two classes of optimization problems: structured optimization, whose solutions exhibit specific structures, and federated learning, which aims to collaboratively learn a model from decentralized data sources. In the first part of this thesis, we characterize the dual correspondence in structured optimization. We further show that this dual correspondence allows us to develop efficient algorithms and design new models for structured optimization. In the second part of this thesis, we propose a dual approach to the federated optimization problem. We demonstrate theoretically and empirically that our approach enjoys better convergence guarantees than other primal-based approaches under specific scenarios. We also explore some application scenarios for structured optimization in federated learning. In the third part of this thesis, we study the problem of evaluating clients' contributions in federated learning. We propose fair and efficient contribution valuation metrics for both horizontal and vertical federated learning, where structured optimization plays a crucial role in our design.

Transporting and evaluating predictive models in different environments 
Farahani, Melika
DOI : 10.14288/1.0418465
URI : http://hdl.handle.net/2429/82622
Degree : Master of Science - MSc 
Graduation Date : 2022-11
Supervisor : Dr. David Poole & Dr. Danica Sutherland

Team harmony before, during, and after COVID-19 
Heyl, Noa Joseph McDonald
DOI : 10.14288/1.0418438
URI : http://hdl.handle.net/2429/82615
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Reid Holmes & Dr. Elisa Baniassad

Better document-level natural language understanding through data-driven applications of discourse theories 
Huber, Patrick
DOI : 10.14288/1.0421296
URI : http://hdl.handle.net/2429/82885
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-11
Supervisor : Dr. Giuseppe Carenini

A discourse constitutes a locally and globally coherent text in which words, clauses and sentences are not solely a sequence of independent statements, but follow a hidden structure, encoding the author's underlying communicative goal(s). As such, the meaning of a discourse as a whole goes beyond the meaning of its individual parts, guided by the latent semantic and pragmatic relationships holding between parts of the document. Clearly falling into the area of Natural Language Understanding (NLU), discourse analysis augments textual inputs with structured representations following linguistic formalisms and frameworks. Annotating documents following these elaborate formalisms has led to the computationally inspired research area of discourse parsing, aiming to generate robust and general discourse annotations for arbitrary documents through automated approaches. With computational discourse parsers having great success at inferring valuable structures and supporting prominent real-world tasks such as sentiment analysis, text classification, and summarization, discourse parsing has been established as a valuable source of structured information. However, a significant limitation preventing the broader application of discourse-inspired approaches, especially in the context of modern deep-learning models, is the lack of available gold-standard data, caused by the tedious and expensive human annotation process. To overcome the prevalent data sparsity issue in the areas of discourse analysis and discourse parsing, it is imperative to find new methods to generate large-scale and high-quality discourse annotations, not relying on the restrictive human annotation process. Along these lines, we present a set of novel computational approaches to (partially) overcome the data sparsity issue by proposing distantly and self-supervised methods to automatically generate large-scale, high-quality discourse annotations in a data-driven manner. In this thesis, we provide detailed insights into our technical contributions and diverse evaluations. Specifically, we show the competitive and complementary nature of our discourse inference approaches to human-annotated discourse information, partially outperforming gold-standard discourse structures on the important task of "inter-domain" discourse parsing. We further elaborate on our generated discourse annotations in regard to their ability to support linguistic theories and downstream tasks, finding that they have direct applications in linguistics and Natural Language Processing (NLP).

Diversity embeddings and the hypergraph sparsest cut 
Jozefiak, Adam Daniel
DOI : 10.14288/1.0418613
URI : http://hdl.handle.net/2429/82677
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Bruce Shepherd

Magic pen: a versatile digital manipulative for learning 
Kianzad, Soheil
DOI : 10.14288/1.0406179
URI : http://hdl.handle.net/2429/80574
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-05
Supervisor : Dr. Karon Maclean

Digital manipulatives such as robots are an opportunity for interactive and engaging learning activities. The addition of haptic and specifically force feedback to digital manipulatives can enrich the learning of science-related concepts by building physical intuition. As a result, learners can design experiments and physically explore them to solve problems they have posed. In my thesis, I present the evolution of the design and evaluation of a versatile digital manipulative – called MagicPen – in a human-centered design context. First, I investigate how force feedback can enable learners to fluidly express their ideas. I identify three core interactions as bases for physically assisted sketching (phasking). Then, I show how using these interactions improves the accuracy of users’ drawings as well as their authority in creative works. In the next phase, I demonstrate the potential benefits of using force feedback in a collaborative learning framework, in a manner that is generalizable beyond the device we invented and lends insight on how haptics can empower digital manipulatives to express advanced concept by means of the behaviour of a virtual avatar and the respective feeling of force feedback. This informs our device’s capability for learning advanced concepts in classroom settings and further considerations for the next iterations of the MagicPen. Based on the findings of how haptic feedback could assist with design and exploration in learning, In the last phase of my thesis, I propose a framework for physically assisted learning (PAL) which links the expression and exploration of an idea. Furthermore, I explain how to instantiate the PAL framework using available technologies and discuss a path forward to a larger vision of physically assisted learning. PAL highlights the role of haptics in future "objects-to-think-with".

AtomicSets.jl: the calculus of support functions in Julia 
Kramer, Mia (Miles Xavian)
DOI : 10.14288/1.0417460
URI : http://hdl.handle.net/2429/82444
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Michael Friedlander

Pedestrian intent estimation through visual attention and time and memory conscious u-shaped networks for training neural radiance fields 
Kuganesan, Abiramy
DOI : 10.14288/1.0421385
URI : http://hdl.handle.net/2429/82947
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Helge Rhodin

Effective techniques of combining information visualization with natural language processing 
Li, Raymond
DOI : 10.14288/1.0406696
URI : http://hdl.handle.net/2429/80914
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Giuseppe Carenini

Reducing the cognitive and temporal costs of software history exploration 
Li, Alison
DOI : 10.14288/1.0417465
URI : http://hdl.handle.net/2429/82438
Degree : Master of Science - MSc
Graduation Date : 2022-11
Supervisor : Dr. Gail Murphy

FASTR : fast approximation of soft tissue in real time 
Liang, Ziheng
DOI : 10.14288/1.0412965
URI : http://hdl.handle.net/2429/81335
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Dinesh Pai

Techniques in learning-based approaches for character animation 
Ling, Hung Yu
DOI : 10.14288/1.0417276 
URI : http://hdl.handle.net/2429/82325
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-11
Supervisor : Dr. Michiel van de Panne

Contemporary computer animation research has benefited substantially from the advancement of deep learning and deep reinforcement learning methods in the past decade. Despite the performance and flexibility of learning-based methods, significant manual effort is still required to tune the training data, algorithms, and environments, especially when computing resources are limited. In this thesis, we develop and evaluate a variety of learning-based methods that enable new skills, and that improve the stability of the learning algorithms and the quality of the synthesized motions. First, we present a framework for learning autoregressive kinematic motion generators and controllers from motion capture data. By disentangling the motion modelling and control tasks, our framework can efficiently synthesize controllable virtual characters by leveraging the strengths of supervised and reinforcement learning. Second, we study the effects of symmetry in learning physics-based locomotion controllers for bipedal characters. We evaluate four possible methods to impose symmetry and show that enforcing symmetry improves the naturalness and task performance of the trained controllers. Third, we explore the role of learning curricula in solving challenging physics-based stepping-stone tasks. The learning is significantly more robust and efficient under a learning curriculum which gradually increases the task difficulty. Finally, we combine simplified models and imitation learning to train brachiation controllers. We show that sparse task objective alone is sufficient for training a controller in an abstracted point-mass brachiation environment. Then, using the point-mass as a reference trajectory for the centre-of-mass, we can learn a control policy for a physics-based 14-link planar articulated gibbon model.

Using neural language models to predict the psychosocial needs of cancer patients
Nunez, John-Jose Andres
DOI : 10.14288/1.0412909
URI : http://hdl.handle.net/2429/81233
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Raymond Ng

Directed multicommodity flows: cut-sufficiency and forbidden relevant minors
Poremba, Joseph Chester
DOI : 10.14288/1.0417583
URI : http://hdl.handle.net/2429/82556
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Bruce Shepherd

Algorithm configuration landscapes: analysis and exploitation
Pushak, Yasha Robert
DOI : 10.14288/1.0413565
URI : http://hdl.handle.net/2429/81496
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-11
Supervisor : Dr. Holger Hoos & Dr. Mark Schmidt

Algorithm designers are regularly faced with the tedious task of finding suitable default values for the parameters that impact the performance of algorithms. Thoroughly evaluating even a single parameter configuration typically requires running the algorithm on a large number of problem instances, which can make the process very slow. To address this problem, many automated algorithm configuration procedures have been proposed. The vast majority of these are based on powerful meta-heuristics with strong diversification mechanisms, thereby ensuring that they sufficiently explore the parameter configuration space. However, despite the prominence of automated algorithm configuration, relatively little is known about the algorithm configuration landscapes searched by these procedures, which relate parameter values to algorithm performance. As a result, while these strong diversification mechanisms make existing configurators robust, it is unclear whether or not they are actually required or simply increase the running time of the configurators. One particularly notable early work in the field showed evidence suggesting that the algorithm configuration landscapes of two optimization algorithms are, in fact, close to uni-modal. However, existing fitness landscape analysis techniques are unable to account for the stochasticity in the performance measurements of algorithms in a statistically principled way, which is a major barrier to their application to studying algorithm configuration scenarios. We address this gap by developing the first statistically principled method for detecting significant deviations from uni-modality in a stochastic landscape. We apply this method, along with other (new and existing) landscape analysis techniques, to a variety of algorithm configuration scenarios arising in automated machine learning (AutoML) and the minimization of the running time of algorithms for solving NP-hard problems. We show that algorithm configuration landscapes are most often highly structured and relatively simple. Using the intuition from this analysis, we develop two prototype algorithm configuration procedures designed for AutoML. We show that the methods make assumptions that are too strong, leading to mixed results. However, we build on this intuition and develop another procedure for the configuration of NP-hard algorithms. Compared to state-of-the-art baselines, we show that our new method often finds similar or better configurations in the same or less time.

Feeling (key)pressed : comparing the ways in which force and self-reports reveal emotion
Reis Guerra, Rubia
DOI : 10.14288/1.0421389
URI : http://hdl.handle.net/2429/82950
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Karon Maclean

Ribbon drawing in VR: brushes and applications
Rosales Ruiz, Enrique Alberto
DOI : 10.14288/1.0412645
URI : http://hdl.handle.net/2429/81128
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-05
Supervisor : Dr. Alla Sheffer

Virtual reality drawing applications let users draw 3D shapes using brushes that form ribbon-shaped, or ruled-surface, strokes. Each ribbon is uniquely defined by its user-specified ruling length, path, and the ruling directions at each point along this path. A collection of these virtual ribbons with proper normal orientations can communicate complex surfaces; thus, artists frequently describe their envisioned 3D surfaces by drawing dense brush strokes that cover the surface of the intended shapes. In this thesis, we analyze these ribbon brushes, and propose ways to expand the scope of their applications and improve their usability. Currently, the practical use of these drawings is limited since most geometry processing algorithms and downstream applications such as 3D printing require manifold meshes. Furthermore, existing brushes use the trajectory of a handheld controller in 3D space as the ribbon path, and compute the ruling directions using a fixed mapping from a specific controller coordinate-frame axis. This fixed mapping requires users to rotate the controller and thus their wrists to change ribbon normal or ruling directions, which requires substantial physical effort to draw even medium complexity ribbons. As people have limited ability to rotate their wrists continuously, the range of ribbon geometries they can comfortably draw with these brushes is limited. We solve these problems by first developing SurfaceBrush, a surfacing method that converts such VR drawings into user-intended manifold free-form 3D surfaces. We then present AdaptiBrush, a ribbon brush system that dramatically extends the space of ribbon geometries users can comfortably draw while enabling them to accurately predict the ribbon shape that a given hand motion produces. Our work expands the range of applications of VR drawing and makes VR drawing a viable alternative to 3D modeling for inexperienced users.

Generalization bounds and size generalization for graph neural networks
Sales, Emmanuel
DOI : 10.14288/1.0417272
URI : http://hdl.handle.net/2429/82330
Degree : Master of Science - MSc 
Graduation Date : 2022-11
Supervisor : Dr. Nick Harvey

Beyond the bulging binder: family-centered design of an information management system for caregivers of children living with health complexity 
Sepehri, Katayoun
DOI : 10.14288/1.0418412
URI : http://hdl.handle.net/2429/82572
Degree : Master of Science - MSc
Graduation Date : 2022-11
Supervisor : Dr. Karon Maclean & Dr. Liisa Holsti (Occupational Science & Occupational Therapy; liisa.holsti@ubc.ca)

Synthesizing multivariate time series using generative adversarial networks with a central discriminator 
Seyfi, Ali
DOI : 10.14288/1.0416398
URI : http://hdl.handle.net/2429/82199
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Raymond Ng

Reinforcement learning in the presence of sensing costs 
Shann, Tzu-Yun
DOI : 10.14288/1.0413129
URI : http://hdl.handle.net/2429/81421
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Leonid Sigal & Dr. Michiel van de Panne

Ufit: interactive attribute driven sewing pattern adjustment 
Shastry, Megha 
DOI : 10.14288/1.0413133
URI : http://hdl.handle.net/2429/81431
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Dinesh Pai

Subspace optimization for machine learning
Shea, Betty
DOI : 10.14288/1.0404490
URI : http://hdl.handle.net/2429/80393
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Mark Schmidt & Dr. Maryam Kamgarpour (EECE)

Parallel locality sensitive hashing for network discovery from time series
Sodol, Svetlana
DOI : 10.14288/1.0406350
URI : http://hdl.handle.net/2429/80726
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Alan Wagner

Learning to get up with deep reinforcement learning
Tao, Tianxin
DOI : 10.14288/1.0412983
URI : http://hdl.handle.net/2429/81297
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Michiel van de Panne

ApproachFinder: real-time perception of potential docking locations for smart wheelchairs
Thukral, Shivam
DOI : 10.14288/1.0406508
URI : http://hdl.handle.net/2429/80782
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Ian Mitchell

Linear information theory and its application to the coded caching problem
Tootooni Mofrad, Amirhossein
DOI : 10.14288/1.0412871
URI : http://hdl.handle.net/2429/81218
Degree : Master of Science – MSc
Graduation Date : 2022-05
Supervisor : Dr. Joel Friedman

Leveraging developer discussions to improve design accessibility
Viviani, Giovanni
DOI : 10.14288/1.0421402
URI : http://hdl.handle.net/2429/82960
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-11
Supervisor : Dr. Gail Murphy

Since the inception of software engineering, the design of a software system has been recognized as one of its most important attributes. A software system’s design determines many of its properties, such as maintainability and performance. One might expect that there is a common and well-established understanding about what software design is and is not. Such an understanding is not evident in the literature, where design has been described in many ways such as large-scale architecture and low-level design patterns, to name just a few. At the same time, an understanding of design is also needed to maintain system properties as changes to the system are made. When developers lose track of the overall design, the system may not conform to its intended properties. Unfortunately, many systems do not have up-to-date design documentation and approaches to recover design often focus on how a system works by extracting structural and behavioural information rather than why it was designed to work like that. In this thesis, we propose an automated approach to extract design information from written discussions between developers. The aim is to make this information accessible to developers, helping them understand the design of a system and make better design decisions. First, we present an interview study we conducted to understand what researchers and practitioners consider as software design. These interviews revealed five recurring topics that can help inform what software design truly represents. We then introduce a classifier able to locate paragraphs in discussions, which we call design points, that pertain to design. Results show that this classifier is able to locate design information with high accuracy even in systems that it was not trained on. We describe a study conducted with software developers that shows that newcomers to a project, when provided with design points relevant to a programming task, are able to interpret and use the design information to consider additional design alternatives. We finally discuss an early exploration into the use of semantic frames to identify useful design points.

Beyond learning curves: understanding stochasiticity and learned solution modes in reinforcement learning
Wilson, Matthew
DOI : 10.14288/1.0420754
URI : http://hdl.handle.net/2429/82765
Degree : Master of Science – MSc
Graduation Date : 2022-11
Supervisor : Dr. Michiel van de Panne

Reinforcement Learning for Legged Robot Locomotion
Xie, Zhaoming
DOI : 10.14288/1.0404507
URI : http://hdl.handle.net/2429/80396
Degree : Doctor of Philosophy - PhD
Graduation Date : 2022-05
Supervisor : Dr. Michiel van de Panne

Deep reinforcement learning (DRL) offers a promising approach for the synthesis of control policies for legged robots locomotion. However, it remains challenging to learn policies that are robust to uncertainty in the real world to put on physical robots or policies that can handle complicated environments. In this thesis, we take several significant steps towards efficiently learning legged locomotion skills with DRL. First, we present a framework to learn feedback policies for a bipedal robotCassie, utilizing rough motion sketches. An iterative design process is then proposed to refine, compress and combine policies for effective sim-to-real transfer. Second, we explore the role of dynamics randomization on a quadrupedal robotLaikago. We demonstrate that with appropriate design choices, dynamics randomization is often not necessary for sim-to-real. We further analyze situations that randomization would become necessary. Third, we propose and analyze multiple curriculum learning approaches to solve the challenging stepping stone tasks for bipedal locomotion. We demonstrate that gradually increasing task difficulties can reliably train policies that solve challenging stepping stone sequences. Finally, we investigate the combination of reinforcement learning and model-based control by training quadrupedal policies using a centroidal model.

ROS-X-Habitat: Bridging the ROS Ecosystem with Embodied AI
Yang, Haoyu
DOI : 10.14288/1.0412641
URI : http://hdl.handle.net/2429/81124
Degree : Master of Science - MSc
Graduation Date : 2022-05
Supervisor : Dr. Ian Mitchell

Detecting viewer-perceived intended vector sketch connectivity
Yin, Jerry
DOI : 10.14288/1.0412642
URI : http://hdl.handle.net/2429/81125
Degree : Master of Science - MSc
Graduation Date : 2022-05
Supervisor : Dr. Alla Sheffer

How subtle design in video games impacts player experience: qualitative studies regarding two video game design features
Yin, Ji Tong
DOI : 10.14288/1.0417303
URI : http://hdl.handle.net/2429/82339
Degree : Master of Science - MSc
Graduation Date : 2022-11
Supervisor : Dr. Robert Xiao

Investigating data-flow reachability questions
Yoo, James 
DOI : 10.14288/1.0421073
URI : http://hdl.handle.net/2429/82822
Degree : Master of Science - MSc
Graduation Date : 2022-11
Supervisor : Dr. Gail Murphy