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
 

Recent Submissions

Item
Facing the Flood: Amphibious Architecture for Flood Resilience in Peguis, Manitoba
(University of Waterloo, 2025-04-17) Holder, Alexa Megan
Floods are Canada's most common and costly natural hazard and flood risks are increasing due to climate change. When floods affect housing, they not only inflict economic and property damage but also displace residents from their homes and affect people’s sense of safety and community. Conventionally, settlements built in flood-prone regions are protected by flood control infrastructure, like levees, dikes, dams, and diversion channels. This infrastructure cannot easily or quickly respond to changing flood conditions and sometimes transfers flood risk rather than mitigating it equitably. We see this in Manitoba’s Interlake Region, where First Nations communities bear a disproportionate burden from water diversion to protect large urban centers. Where conventional solutions have failed and new tools are urgently needed, amphibious construction can provide an option for mitigation. Amphibious structures sit on dry land when water levels are normal, like an ordinary building. However, there is a buoyancy system, allowing the structure to float on the water in a flood. Vertical guidance, often posts, holds the building in place laterally while floating. When flood waters recede the building returns to its original position undamaged. Inspired by community members who wish to stay on their land despite flood risks, this thesis proposes amphibious architecture for a site in the First Nations Community of Peguis, Manitoba. Relocated to flood-prone land after a fraudulent land transfer in 1907, the community experiences chronic flooding. They faced several floods in the last two decades and had their worst flood on record in 2022, emphasizing the urgency of providing solutions for residents. This thesis examines this history and a specific site in Peguis, identifying key considerations for implementing amphibious architecture there. Then, it assesses existing amphibious architecture precedents, looking at how these projects address common challenges. Drawing insights from this analysis, this thesis proposes a prototype design for the site in Peguis.
Item
FPGA-Accelerated Deep Learning for Denoising Low-Dose PET Scans
(University of Waterloo, 2025-04-17) Dao, Eric-Khang Tan
Positron Emission Tomography (PET) is an essential imaging technique used in clinical settings for diagnosing conditions such as cancer and neurological disorders; however, its dependence on radiopharmaceuticals poses potential radiation exposure risks. Lowering the administered dose can help improve patient safety but results in imagery with reduced Signal-to-Noise Ratio (SNR), impacting diagnostic accuracy. The trade-off between minimizing radiation exposure and maintaining image quality remains a key challenge in PET imaging. Recently, deep learning-based denoising techniques, such as Denoising Convolutional Neural Network (DnCNN), have proven effective in restoring noisy images to standard quality. Traditional implementations relying on CPUs and GPUs are often constrained by high power consumption and hardware overhead, limiting feasibility in edge-compute applications. To address these challenges, this thesis explores FPGA-based acceleration for PET image denoising. A dataset is constructed using PET scans from 10 Alzheimer’s disease patients from the ADNI database, with only 0.5% of the original radiotracer dose used. A software-based implementation is developed using a proposed U-Net-like architecture, then ported to an FPGA using OpenVINO and Intel’s FPGA AI Suite for hardware emulation. Experimental results show the FPGA implementation offers a 77% improvement in performance-to-watt ratio compared to the GPU-based solution, and a 2x reduction in latency compared to the CPU-based solution.
Item
Queer Arrival: Uncovering the Spatial Narratives of QTPOC Newcomers in Toronto
(University of Waterloo, 2025-04-17) Liao, Simon
Toronto’s urban landscape is continuously shaped by immigrants, queer, and marginalized communities. Historically, immigrants have established ethnic “Arrival Cities” to foster mutual support, and queer communities have carved out queer spaces like the Church-Wellesley Village to cultivate safety, belonging, and visibility. Positioned at the intersectionality of marginalized identities, Queer and Trans People of Colour (QTPOC) newcomers are also actively contributing to the evolution of the urban landscape, giving rise to a new spatial typology – the “Queer Arrival City”. Existing research on Arrival Cities and queer enclaves remains constrained within narrow conceptual boundaries, overlooking the broader spectrum of urban arrival. Arrival Cities are typically examined through an ethnic minority lens, focusing on neighbourhood dynamics, while queer enclaves are studied predominantly from a white, middle-class gay male perspective. These approaches neglect the intersectionality of race, gender, sexuality, and class in the production of diasporic spaces, leaving QTPOC newcomers underrepresented in both academic and public spheres. This thesis addresses these gaps by uncovering the spatial narratives of Toronto’s QTPOC newcomers in constructing their “Queer Arrival City”. It specifically examines how QTPOC newcomers navigate Toronto’s built environment and the role of the Church-Wellesley Village in their migration. Furthermore, it explores the design of a public space that materializes QTPOC newcomers’ spatial narratives as a place of belonging and visibility. This research employs a Queer of Colour Methodology (QOCM) integrated with Participatory Action Research (PAR) to foreground intersectionality and actively engage QTPOC newcomers in both the research and design process. Drawing on qualitative and quantitative data from three phases of community engagement - surveys, interviews, and focus groups, this thesis introduces the novel “Queer Arrival” framework, encompassing both infrastructural and individualized spatial typologies, while articulating a collective “Queer Diaspora Spatial Consciousness” in inhabiting public space. The research culminates in a design proposal shaped by the active contribution and lived experiences of QTPOC newcomers. Ultimately, by positioning QTPOC newcomers as the primary holders of knowledge production, this thesis fosters an inclusive, community-driven research environment and design process, while prioritizing QTPOC newcomers’ empowerment and agency in shaping their future built environment.
Item
Grafting of Starch Nanoparticles with Polymers
(University of Waterloo, 2025-04-17) Fernandez, Joanne
As a biocompatible and biodegradable polysaccharide, starch has sparked significant interest for various industrial applications, but its poor mechanical properties limit its uses without chemical or physical modification. The work reported herein concerns the development of synthetic techniques to modify starch by graft polymerization via cerium (IV) activation. Starch nanoparticles (SNPs) were modified with acrylic acid (AA) in water under acidic conditions via activation with cerium (IV) in combination with potassium persulfate (KPS). The reactions were conducted with either the as-supplied SNPs containing glyoxal, or after purification (without glyoxal), for different target molar substitution (MS) values. A novel purification protocol using methanol extraction and centrifugation was implemented to purify the samples. This method proved to be selective to isolate the poly(acrylic acid) (PAA) homopolymer contaminant from the starch-g-PAA copolymer, and more reliable than the gravimetric analysis methods reported in the literature. The starch-g-PAA copolymers were characterized by dynamic light scattering (DLS), and degradation of the starch substrate allowed the determination of the molar mass of the PAA side chains via gel permeation chromatography (GPC) analysis. In the presence of aldehydes the rate of polymerization of AA increased significantly (by > 37 %), and the highest grafting efficiencies were obtained for glyoxal and butyraldehyde. The combination of cerium (IV) with glyoxal and KPS resulted in the highest polymerization rate and grafting efficiency. Increasing the glyoxal concentration also increased the rate of monomer conversion and the grafting efficiency. The increased rate of polymerization provided further insight into the grafting mechanism, as it was discovered that esterification reactions between starch and PAA also contributed significantly to the grafting process, particularly at longer reaction times. In the presence of aldehydes, the production of large amounts of PAA homopolymer resulted in esterification dominating the grafting process. Model reactions involving direct coupling of linear PAA samples with starch were investigated. All the reactions were characterized by high coupling efficiencies for a target MS = 3, and higher molar mass PAA samples (30 and 250 kDa) coupled faster than a lower molar mass sample (1.8 kDa), as expected in terms of reaction probabilities. The importance of esterification was also confirmed with model reactions using 2-hydroxyethyl acrylate, a monomer not containing a free carboxylic acid functional group, which yielded notably lower grafting efficiencies. Overall, the grafting mechanism for starch and acrylic acid promoted by cerium (IV) therefore appears more complex than described previously, particularly in the presence of aldehydes: The high overall grafting efficiencies observed result from two distinct reactions occurring concurrently, namely grafting via cerium (IV) activation, as well as the esterification of free PAA homopolymer. The additional insight gained for these reactions was possible due to the newly developed purification protocol, used in combination with NMR spectroscopy analysis, which provided detailed composition data for the different sample fractions and a better understanding of the grafting mechanism. Furthermore, preliminary results were obtained for starch modified with acrylonitrile and cerium (IV) in water under acidic conditions. Extraction of the polyacrylonitrile (PAN) homopolymer component was more difficult due to its solubility characteristics, but mixtures of dimethylacetamide with water (up to 10 % by volume) provided consistent results. High grafting efficiencies (> 67 %) were obtained for the starch-g-PAN copolymers, and characterization of the products was performed by Fourier transform-infrared spectroscopy, DLS, GPC, and atomic force spectroscopy. Hydrolysis of the starch substrate yielded hollow PAN shells or spheres, depending on the MS level of the copolymer, with potential applications in nanoencapsulation.
Item
Language Model Inference on FPGA with Integer-only Operations
(University of Waterloo, 2025-04-17) Bekmyrza, Marat
Large Language Models (LLMs) are currently dominating the field of Artificial Intelligence (AI) applications, but their integration for edge computing purposes is rather limited due to computational complexity and power consumption. This thesis addresses this challenge by investigating the integer-only acceleration of transformer models on FPGAs, focusing on the BERT architecture. We demonstrate that by removing the floating-point operations from the inference pipeline, especially from non-linear functions like GELU, Softmax, and Layer Normalization, we can improve the performance without sacrificing accuracy. Our pipelined, batched architecture processes multiple sequences in parallel and optimizes the FPGA resources. We achieve a 2.6x throughput improvement over a single-sequence inference and at least 10x speedup over the offloading to CPU approach. The results of the experiments show that our implementation has comparable accuracy to the floating-point models for the GLUE benchmark tasks with INT8 quantization. These findings reveal that integer-only transformer inference on FPGAs is a feasible way of implementing complex language models on resource-limited edge devices, with the potential for new privacy-conscious, low-latency AI applications.