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 ,
    Robust optimisation for sequential investment problems: an application to climate adaptation in the Mississippi River Basin
    (Taylor & Francis, 2026-03) Wu, Zhenggao; Dimitrov, Stanko; Pavlin, Michael
    Effectively adapting to climate change requires long-term investment strategies informed by climate forecasts. This paper presents methods and a case study assessing how an investor would approach sequential land investments in the Mississippi River Basin (MRB) using 32 climate models under two emission scenarios (RCP4.5 and RCP8.5). Each model produces projected farmland values, which are used in a robust optimisation framework to identify optimal investment policies under varying levels of conservatism-reflecting the degree to which worst-case outcomes are considered. The model is linearised and scalable across long horizons and asset sets. The case study spans 2023-2090 and uses regression-based projections of land values. Robust investment strategies are derived for each climate model and scenario. Results show that as conservatism increases, investment becomes more geographically constrained, with significant variation in optimal regions. While all climate forecasts predict warming, the regions benefiting from such changes vary by model and scenario. Investment patterns range from concentration in specific areas to broad diversification or complete withdrawal. The analysis underscores substantial disagreements among climate models, which critically affect spatial investment decisions and highlight the importance of robust planning under climate uncertainty.
  • Thumbnail Image
    Item type: Item ,
    Analysis of Neural Networks with Physics Applications
    (University of Waterloo, 2026-03-30) Mohamed, Ahmed
    This thesis investigates core aspects of machine learning, spanning foundational studies on generalization phenomena in neural networks, novel architectural strategies for enhancing representation learning and classification performance, and high-accuracy predictive and inverse modeling of emerging nanoelectronic devices. Together, these studies highlight the significance of data and model structure, the impact of nonlinearity, and the potential of interpretable, generalizable machine learning methods for scientific and engineering applications. For generalization in neural networks, the thesis focuses on the phenomenon of grokking, a delayed generalization effect where models initially overfit but eventually learn to generalize well after extended training. Through a series of interconnected studies, this work proposes insights and practical tools to diagnose, forecast, and enhance generalization in modern machine learning systems. The first part of the thesis examines grokking in modular arithmetic tasks, revealing how dropout-induced variance, embedding similarity, activation sparsity, and weight entropy evolve across training, and hence introduces diagnostic metrics to capture phase transitions between memorization and generalization. Further analysis shows that nonlinearity, network depth, and symmetry in data collectively modulate grokking behavior, linking model architecture to its capacity for structured generalization. Next, the thesis introduces a Branched Variational Autoencoder (BVAE), a hybrid architecture that integrates generative and discriminative objectives. By shaping latent representations through a supervised branch, the BVAE achieves improved class separability and interpretability on benchmark datasets, illustrating the potential of structured latent shaping for semi-supervised learning. Finally, the research extends to scientific machine learning, demonstrating how neural and ensemble models as Random Forests can accelerate the modeling and inverse design of Carbon Nanotube Tunnel Field-Effect Transistors (CNT TFETs). By coupling physical insights with machine learning interpretability techniques, this work bridges the gap between theoretical ML and real-world scientific applications.
  • Thumbnail Image
    Item type: Item ,
    Linear Scala is All You Need for Safe Static Memory and Alias Management
    (University of Waterloo, 2026-03-30) Pashaeehir, Amirhossein
    Rust has become one of the most popular languages for systems programming. This popularity is largely driven by Rust's ability to provide safe static memory management without garbage collection, eliminating GC-induced pauses and runtime overhead that can be difficult to predict and control. In addition, Rust enforces alias and mutability control through its ownership and borrowing discipline, enabling features such as fearless concurrency and stronger compiler optimizations while preserving memory safety. Scala Native brings Scala to systems-level targets by compiling to LLVM IR. However, it still relies on third-party garbage collectors for memory management and does not provide Rust-style static guarantees for safe memory management, aliasing, and mutability control. This thesis presents imem, a library that brings Rust-inspired ownership and borrow checking to Scala, and Scinear, a minimal compiler plugin that adds linear types to Scala and integrates them with capture checking and polymorphism. imem proves that, given Scala's type system, linearity is the only missing ingredient needed to implement most of Rust's ownership and borrowing discipline as a library rather than a dedicated language feature. imem provides linear Box values and immutable and mutable references, enforces ownership rules, and statically controls aliasing and mutability, following the Stacked Borrows model. In addition, imem offers optional runtime verification to detect potential safety violations when users apply workarounds to the static rules. To demonstrate practicality, the thesis develops a safe linked-list case study and compares imem against Rust, vanilla Scala, and linear Scala. The evaluation shows that imem matches Rust's level of expressiveness, so it can support list operations and iterators alongside statically enforcing ownership rules and controlling mutability.
  • Thumbnail Image
    Item type: Item ,
    Considerations for the Design of a UD-NCF Composite Energy Absorbing Structure for Frontal and Oblique Crush Loading
    (University of Waterloo, 2026-03-27) Huang, Ningwei
    Growing concerns regarding climate change have prompted national and international regulatory agencies to implement increasingly strict regulations aimed at reducing carbon dioxide (CO₂) emissions. These regulations have driven automotive manufacturers to place greater emphasis on sustainability and improved fuel efficiency in vehicle development. Owing to their high specific strength and stiffness, and superior energy absorption capability, carbon fiber-reinforced plastic (CFRP) composites are considered promising lightweight materials for vehicle frontal crash structures. Their widespread adoption in the automotive industry was previously limited due to high manufacturing costs and challenges in accurately predicting their response under impact loading. However, CFRP components manufactured via high-pressure resin transfer (HP-RTM) with highly reactive resins enable reduced production cycle times and, thus, adoption in automotive structures. Unidirectional non-crimp fabric (UD-NCF) reinforcements offer further advantages, including reduced manufacturing costs, high in-plane mechanical properties, and enhanced design flexibility. To meet safety requirements, vehicle structures must be designed to effectively absorb energy under various impact conditions to protect the occupants from injury. Previous studies have primarily focused on evaluating the impact performance and energy absorption characteristics of CFRP composite components under axial loading. Few studies have investigated the effects of oblique loading on the crush performance of composite structures and they are mainly restricted to closed-profile tubes, which are difficult to manufacture using liquid composite molding technologies such as HP-RTM. To date, the crush performance of UD-NCF composite components under oblique loading has not been examined. Therefore, this thesis aims to design a UD-NCF composite frontal crush component capable of achieving progressive energy absorption under both axial and 30-degree oblique loading conditions. The design is limited to adhesively bonded double channel components as they can be readily fabricated using HP-RTM processes, while the scope of the study is intended to address several considerations for this design concept. Firstly, the energy absorption capability and failure modes of UD-NCF composite single and double hat channel specimens with [0/±45/90]s and [±45/02]s stacking sequences under quasi-static oblique (i.e., 30-degree off-axis) loading were experimentally investigated to provide data for validation of an impact simulation model. For the single hat channel, specimens with a [0/±45/90]s layup achieved 0.78% higher total energy absorption and 11.2% higher specific energy absorption (SEA) than specimens with a [±45/02]s layup. Specimens with both stacking sequences exhibited lamina bending during the initial crushing stage, followed by premature failure. For the adhesively bonded double hat channel, the [±45/02]s specimens yielded 6.4% higher total energy absorption and 9.95% higher SEA than the [0/±45/90]s specimens due to their higher axial stiffness. The double hat channel configuration demonstrated significant improved crush stability than single hat channels throughout the loading process, regardless of stacking sequences. Secondly, computer-aided engineering (CAE) impact simulation models were developed to predict the energy absorption capability of UD-NCF composite channels under quasi-static and dynamic crushing conditions. Simulation models for both single and double hat channel specimens were validated against the performed oblique crushing experiments and exiting axial crush test data from the literature. The results showed that the CAE impact simulation model accurately predicted the crush performance for both single and double hat channel specimens under dynamic axial loading, while having reduced accuracy under quasi-static loading conditions. Lastly, the influence of channel cross-sectional geometry and laminate stacking sequence on energy absorpiton capacility of the UD-NCF composite channels under dynamic oblique loading was investigated using the validated simlation models. Single and adhesively bonded double channels with five distinct geometries and six stacking sequences were considered in the study. All single channel geometries with a [0/±45/90]s stacking sequence exhibited similar SEA, which was the case for both axial and oblique dynamic loading. Under oblique loading, all double channel geometries with [0/±45/90]s and [0/±45/90/±30]s stacking sequences exhibited premature failure. The hat channel geometry consistently demonstrated stable progressive crushing, whereas the other geometries considered showed greater sensitivity to stacking sequence and loading angle. Across all stacking sequences considered, only channels with a [±45/02]s stacking sequence achieved stable crushing under both axial and oblique loading, while also providing the highest SEA values. The double hat channel with the [±45/02]s stacking sequence was identified as the most promising configuration for subsequent frontal crush structure design. Overall, this represents the first comprehensive assessment of the crush performance of UD-NCF composite components under oblique loading conditions. These findings contribute practical design guidelines for the future development of lightweight UD-NCF frontal crush structures in vehicles.
  • Thumbnail Image
    Item type: Item ,
    From Asymptotic to Finite-Size Security in Decoy-State Quantum Key Distribution
    (University of Waterloo, 2026-03-24) Kamin, Lars
    Quantum Key Distribution (QKD) promises information-theoretic security, yet bridging the gap between theoretical proofs and practical implementations, specifically those operating with finite resources and imperfect devices against general coherent attacks, remains a critical challenge. This thesis develops a spectrum of efficient security proof techniques within the composable security framework, calculating key rates for both fixed- and variable-length protocols while accounting for realistic imperfections. We begin by addressing detection setups through an extension of a squashing map, the flag-state squasher, used for reducing the infinite-dimensional Hilbert spaces of optical elements to finite dimensions. This extension accommodates arbitrary passive linear optical setups while allowing for the inclusion of detection inefficiencies and dark counts in the security analysis. Subsequently, we advance the analysis of decoy-state protocols and introduce two major improvements. First, we reformulate the decoy-state analysis to recover no-decoy key rates, tightening the optimization. Second, we derive a unified framework that performs the key rate optimization and decoy analysis in a single step. This enables the bounding of the relevant entropies with arbitrary precision in the finite-size regime and successfully recovers the Devetak-Winter formula in the asymptotic limit. Furthermore, we improve the security analysis for generic QKD protocols against independent and identically distributed (IID) collective attacks. Our refined analysis yields finite-size corrections proportional to detected rather than transmitted signals and, by developing sharper concentration inequalities, achieves significantly improved finite-size scaling. Finally, leveraging the marginal constrained entropy accumulation theorem (MEAT), we establish a flexible numerical Rényi security framework against coherent attacks for both fixed- and variable-length protocols. This approach consistently outperforms existing reference proof techniques, including those based on entropic uncertainty relations, providing significantly higher key rates for both qubit and practically relevant decoy-state protocols. Moreover, we present finite-size key rates for generic QKD protocols accounting for realistic intensity and phase imperfections. Overall, this thesis provides the necessary theoretical framework to bridge the gap between idealized models and experimental reality, offering a scalable path toward secure quantum communication under realistic conditions, as demonstrated by the application of these techniques in experimental collaborations.