UWSpace

UWSpace is the University of Waterloo’s institutional repository for the free, secure, and long-term home of research produced by faculty, students, and staff.

Depositing Theses/Dissertations or Research to UWSpace

Are you a Graduate Student depositing your thesis to UWSpace? See our Thesis Deposit Help and UWSpace Thesis FAQ pages to learn more.

Are you a Faculty or Staff member depositing research to UWSpace? See our Waterloo Research Deposit Help and Self-Archiving pages to learn more.

Photo by Waterloo staff

Communities in UWSpace

Select a community to browse its collections.

Recent Submissions

  • Thumbnail Image
    Item type: Item ,
    Soft Labels for Training and Evaluating Semantic Segmentation Models
    (University of Waterloo, 2026-04-17) Jamali, Nima
    Despite notable advances in network architectures and representation learning, most semantic segmentation pipelines continue to rely on hard ground-truth labels and evaluation metrics originally designed for binary masks. This assumption is misaligned with real-world data, where object boundaries are often ambiguous, annotations are noisy, spatial downsampling aggregates multiple semantic classes, and uncertainty is frequently encoded through void labels that are ignored during training and evaluation. As a result, both learning objectives and evaluation criteria do not faithfully reflect the underlying semantic structure and uncertainty of the data. In this thesis, we study semantic segmentation in settings where the ground-truth labels are not necessarily hard, but instead may be uncertain. We refer to this formulation as soft-label semantic segmentation. We treat this problem in a unified, end-to-end manner encompassing label generation, training, and evaluation. To generate soft labels, we propose a geometry-aware downsampling strategy called Weighted Average Pooling (WAP ) for semantic segmentation masks. WAP produces smooth and probabilistically valid soft labels at arbitrary resolutions by constructing spatially varying weights based on geometric relationships and spatial proximity. As a result, the generated soft labels are resolution-agnostic, preserve the underlying probabilistic structure of the annotations, and avoid artifacts commonly introduced by conventional downsampling methods. In addition, this work motivates the need for evaluation metrics that operate directly on probabilistic segmentation outputs. To this end, we introduce several principled relaxations of the soft intersection-over-union (soft IoU) metric that provide faithful extensions of standard IoU to soft-label settings. We further introduce void replacement strategies that assign soft class distributions to void pixels based on spatial context, enabling uncertain regions such as ambiguous boundaries and thin structures to be incorporated directly into the supervision signal. Extensive experiments on the PASCAL VOC 2012 dataset demonstrate that WAP produces more faithful soft labels than conventional approaches, particularly in scenarios involving thin structures and complex spatial arrangements. The proposed soft IoU relaxations offer improved interpretability and better alignment with hard-label evaluation, while the void replacement strategies perform comparably to hard-label baselines, indicating that incorporating soft supervision in uncertain regions does not compromise segmentation quality. Together, these contributions establish a principled framework for generating, training, and evaluating soft labels in semantic segmentation.
  • Thumbnail Image
    Item type: Item ,
    Targeted Gene Delivery to Astrocytes Using Intelligent Phagemid-Assembled Gene Expression (iPhAGE) Technology for Neuronal Regeneration
    (University of Waterloo, 2026-04-17) Ffrench, Anna
    Neurodegenerative diseases, stroke, and brain injuries lead to progressive neuronal loss, with limited treatment options focused mainly on symptom management rather than regeneration. A major challenge in developing effective therapies is the lack of safe and efficient gene delivery systems capable of targeting cells in the central nervous system. This project addresses this gap by developing a novel gene therapy approach using the Intelligent Phagemid-Assembled Gene Expression (iPhAGE) platform, an innovative bacteriophage-based system designed to deliver reprogramming transcription factors to astrocytes. The iPhAGE system, derived from M13 miniphagemids, offers high transfection efficiency, large cargo capacity, low immunogenicity, and blood-brain barrier permeability, making it an ideal tool for targeted gene delivery. This research involves constructing plasmids carrying the NeuroD1 gene under astrocyte-specific and universal promoters, producing and characterizing miniphagemids, and evaluating their transfection efficiency in resting and activated A7 cells. Neuronal reprogramming is assessed through PSA-NCAM neuronal marker analysis, a marker associated with early neuronal differentiation, used to evaluate phenotypic changes following NeuroD1 delivery. Key methodologies include designing and validating precursor and helper plasmids, producing miniphagemids, and characterizing them using restriction digestion, MADLS, and transfection assays. The outcome of this project was the establishment and initial evaluation of the iPhAGE-based gene delivery framework for astrocyte-to-neuron reprogramming. This work assesses the main components of the iPhAGE system, including vector construction, miniphagemid production, and astrocyte reprogramming using neuronal markers, therefore establishing a groundwork for future optimization.
  • Thumbnail Image
    Item type: Item ,
    Advancing the Conditional Source-Term Estimation (CSE) Framework for Turbulent Combustion Modeling and Application to Alternative Aviation Fuels
    (University of Waterloo, 2026-04-16) Mahdipour Dilmaghani, Amir Hossein
    The aviation sector faces increasing pressure to reduce greenhouse gas emissions and pollutant formation while meeting projected growth in global air traffic. Alternative aviation fuels (AAFs), including oxygenated fuels and hydrogen, represent promising pathways toward cleaner combustion, provided that their complex chemistry--turbulence interaction can be accurately predicted. Numerical simulation plays a critical role in this effort, but its reliability depends on the availability of combustion models that are both computationally efficient and sufficiently accurate for complex fuels and turbulent flow regimes. This thesis focuses on the development, assessment, and extension of the Conditional Source-term Estimation (CSE) model for the simulation of turbulent non-premixed flames, with particular emphasis on alternative aviation fuels. First, several CSE formulations are systematically evaluated, including traditional Tikhonov-regularized inversion and a Bernstein-polynomial-based approach, with the aim of improving numerical stability, accuracy, and computational efficiency. Next, a new CSE framework incorporating direct integration of detailed chemical kinetics is introduced, eliminating the reliance on pre-tabulated chemistry and enabling more robust predictions for fuels and conditions that are not well represented by conventional chemistry manifolds. The proposed developments are validated against well-documented laboratory-scale turbulent jet flames, including methane, dimethyl ether (DME), and hydrogen flames, covering different fuels and flow scenarios. The effects of differential diffusion are incorporated into the CSE-direct chemistry framework for hydrogen flames, addressing a key limitation of previous CSE implementations. An adaptive and automated CSE ensemble definition strategy is also presented and tested in large eddy simulation (LES) to capture local extinction dynamics. Overall, this work advances the CSE methodology by enhancing its flexibility, physical fidelity, and applicability to alternative aviation fuels. The results demonstrate that CSE with direct chemistry integration can achieve accurate turbulence–chemistry interaction predictions while maintaining computational efficiency, thereby providing a viable modeling framework for future simulations of advanced combustion systems relevant to sustainable aviation.
  • Thumbnail Image
    Item type: Item ,
    Practical Distributed Key Generation and Signatures
    (University of Waterloo, 2026-04-16) Komlo, Chelsea
    Threshold schemes are a critical cryptographic primitive that allows a set of n total parties and a threshold of at least tparties to collaborate to jointly perform some function, such as generating key material or issuing a digital signature. Threshold schemes allow for improved robustness in the case of failure, and distribute trust among many parties. In particular, the security of the scheme assumes t−1 players are corrupted, and so can deviate arbitrarily from the protocol. The security of the scheme ensures that in spite of a subset of corrupted players, the scheme can provide important properties such as robustness, unforgeability, or indistiguishability from some target distribution. In this work, we examine the special cases of distributed key generation and threshold signing. In particular, we present constructions that optimize for considerations that are important to implementations in practice. Such considerations include simplicity, network round efficiency, computational and bandwidth efficiency, and low use of broadcast channels. Firstly, we present FROST, a Flexible Round-Optimized Schnorr Threshold signature scheme. FROST improves upon prior threshold Schnorr signature schemes in that signatures can be generated with only two network rounds among participants, while remaining secure against concurrent adversaries. We show that FROST is secure under the Algebraic One-More Discrete Logarithm (ℓ-AOMDL) assumption in the Random Oracle Model (ROM). Secondly, we present Storm, a simplified three-round distributed key generation protocol (DKG). Storm presents a simplified alternative to prior DKGs with a similar security model, assuming the Discrete Logarithm Problem (DLP) is hard, and provides a generic construction that may be applicable beyond discrete-logarithm assumptions. Finally, we present Arctic, a two-round deterministic threshold Schnorr signature scheme. Arctic allows signers to remain stateless, with the exception of persisting state of their long-lived signing keys. Arctic requires a slightly weaker trust model in that it assumes the majority of signers are honest, but shows improved efficiency over alternative deterministic threshold Schnorr signature schemes for small signing coalitions (fewer than 25 signers). We show that Arctic is secure assuming DLP in the ROM.
  • Thumbnail Image
    Item type: Item ,
    Development of Cross-Conjugated Polymers for Sensing Applications
    (University of Waterloo, 2026-04-16) Zhao, Naixin
    Conjugated polymers are an important class of materials for electronic applications. Compared to conventional inorganic semiconductors, they offer mechanical flexibility, solution processability, and tunable electronic properties. Recently, the research interest in polymer-based sensing technology has grown considerably due to the increasing demand from emerging fields such as the Internet of Things (IoT), smart packaging, and healthcare electronics. Sensors based on conjugated polymers have demonstrated promising performance towards various stimuli such as liquid chemicals, gaseous compounds, and temperature. However, they still suffer from several limitations, including insufficient stability, reversibility, and manufacturing challenges. This thesis aims to address these issues through the exploration of novel material designs based on cross-conjugated building blocks, which have received less research interest compared to linear conjugated structures due to their inherent lower carrier mobility. For sensing applications, their unique ability to transform into a linear conjugated structure under specific stimuli could be beneficial for enhanced sensor sensitivity and selectivity. Additional design strategies are employed for enhanced sensing performance, including the incorporation of hydrogen-bonding sites for reversible sensing and the development of intrinsically conductive polymers to eliminate the need for external dopants, potentially improving device stability. In the first part of the study, dihydropiperazine (DHP) is chosen as the target cross-conjugated building block. A novel building block bisindolin-dihydropiperazine (IDHP) is developed and further copolymerized with a thienothiophene (TT) unit to constitute the cross-conjugated polymer, PIDHPTT. IDHP monomer exists as a cross-conjugated lactam but converts to a conjugated lactim form within the polymer. Neighboring DHP units in the lactim form facilitate this process through π-bridges, demonstrating a vinylogous effect, which has previously only been observed in small molecules. The OH groups in the lactim DHP interact more strongly with fluoride ions (F-) than other halides (Cl-, Br-). A water-gated organic field effect transistor (WGOFET) sensor based on PIDHPTT shows excellent sensitivity (LOD = 0.28 μM) and selectivity for fluoride ions over other halide ions, in addition to excellent reversibility and high stability in ambient and aqueous environments, demonstrating the potential of this polymer design for aqueous chemical sensing applications. Next, two thiophene-flanked DHP-based polymers, PTDbT-ET and PTDbT-T, are developed and synthesized with the eco-friendly DArP method. Incorporation of tri-ethylene glycol (TEG) side chain significantly raises their HOMO energy level higher than the ambient oxygen oxidative potential, enabling spontaneous doping by oxygen gas in the presence of moisture. Due to the higher abundance of TEG groups in PTDbT-T, it possesses a larger energy trap between its HOMO and oxygen oxidative potential, forming a more stable charge transfer complex (CTC) and can maintain its conductivity by storing in a moisture-free environment. When tested toward volatile organic compound (VOC) gaseous analytes, PTDbT-T-based chemiresistive sensors demonstrate excellent repeatability and stability, in addition to high sensitivity and selectivity to ethanol (LOD = 3.07 ppm) over other alcohol species, demonstrating the potential of this alternative strategy to develop dopant-free conductive polymers for chemiresistive gas sensor applications. In the second part of the study, a novel cross-conjugated polymer and the first polymeric analogue of a quinhydrone-like charge-transfer complex with intrinsic conductivity, poly(3,4-dihydroxythiophene-alt-thiophene-3,4-dione) (P(HOT-DOT)), is designed and synthesized. The ammonia-coordinated polymer P3 generates a perfectly balanced 1:1 donor-acceptor architecture that promotes self-doping and stabilizes polarons with spontaneous air oxidation. The polymer exhibits a narrow bandgap, broad near-infrared absorption, and high intrinsic conductivity (∼0.29 S cm-1), enabled by an ultrasmall π-π stacking distance (3.25 Å) despite its cross-conjugated backbone. Flexible temperature sensors fabricated from P3 show high stability, rare positive temperature coefficient (PTC) behavior, and reproducible and linear thermal responses over multiple cycles (TCR = 0.113 ± 0.00045%/°C). Ongoing and future studies of this material should focus on expanding other basic coordination groups for higher material stability and targeting unique electronic properties for high-performance organic electronics applications.