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 ,
    Evaluating and Improving Generative Models with Feedback
    (University of Waterloo, 2026-11-05) Jiang, Dongfu
    Generative models can produce many candidate outputs for a request, but deciding which output is good, and what a model should learn next, requires feedback, a judgment of how well an output serves its task. This thesis studies how to obtain such feedback and how to use it, for evaluating generated outputs and for improving the models that produce them. The two uses place different demands on feedback. For evaluation, feedback has to agree with human judgment and explain its judgment. For training, it has to stay reliable while a model is optimized against it. The first part studies feedback for evaluation. For text, I show that feedback can be learned without a reference output, by training an evaluator on error analyses so that it locates and explains each error in a response, and its scores agree with human ratings nearly as well as GPT-4 used as a judge. For images and videos, I collect pairwise human votes on an open platform, use them to rank generative models, and release them as a benchmark on which the best multimodal judge tested agrees with people about half the time, so human preferences remain the reference for this content. The second part studies feedback for improving models, where agreement with people on fixed outputs is not enough. A policy trained against the learned text evaluator learned to refuse rather than answer, raising its reward while its accuracy fell. For code, I synthesize tests for tens of thousands of programming problems and turn execution results into preferences and rewards, which improve both Best-of-N selection and reinforcement learning. For agents that use tools over several turns, I build a reinforcement-learning framework that runs tools inside the training loop, train agents in six domains, and examine two choices in the feedback they receive. Rewarding tool use raises tool calls without necessarily raising task performance, and how much of a tool's output the policy reads can change the result. Across these studies, feedback that evaluates fixed outputs well can behave differently once a model is trained against it, so a training reward needs a measure it does not see, tracked while the model is optimized.
  • Thumbnail Image
    Item type: Item ,
    Determining the role of fluid flow on lens biomechanics
    (University of Waterloo, 2026-11-05) Gorla, Shilpa
    The vertebrate crystalline lens is a highly specialized, avascular, and transparent structure whose primary function is to change shape to refract incident light onto the retina. With increasing age, however, the lens undergoes progressive compaction of the fiber cells and crosslinking of the structural proteins, causing it to lose its elasticity. This age-dependent decline in lens ability to change shape is called presbyopia, an ocular condition affecting approximately 2 billion people worldwide. The lens of the eye relies on a specialized microcirculatory system (MCS) to maintain transparency, nutrient supply, and volume homeostasis in the absence of a blood supply. Given that the MCS plays an essential role in volume homeostasis and there is an age-related decline in the function of MCS, modulating the components of the MCS with agonists and antagonists offers a powerful approach to understanding the precise mechanism of presbyopia and identifying novel therapeutic targets for presbyopia. Using chickens, which can deform their lenses to induce shape changes and naturally develop age-related presbyopia, this study investigated the following objectives: 1. To determine the effect of modulators of aquaporin 0 (AQP0) and sodium-potassium ATPase (NKA) pumps on the lens biomechanics. 2. To determine whether TRPV1 and TRPV4 are present in the chicken crystalline lens and how modulating them affects lens biomechanics and protein expression. 3. To determine the temperature-dependent role of the lens microcirculatory system (MCS) in regulating lens biomechanics. We found that the chicken model is a robust model for investigating lens biomechanical regulation by the MCS. At room temperature, modulating AQP0, water channels altered lens hydrostatic pressure, whereas NKA modulation affected biomechanics only when ionic or volume homeostasis was disturbed. Western blot analysis confirmed the presence of protein expression of both mechanosensitive TRPV1 and TRPV4 channels in chicken lens tissue, mirroring mammalian expression patterns. TRPV4 modulation altered lens biomechanics exclusively at the physiological temperatures, whereas AQP0 and NKA modulation produced no biomechanical effect at the physiological temperature, highlighting temperature-dependent regulation of the lens microcirculatory system.
  • Thumbnail Image
    Item type: Item ,
    Universality between experiment and simulation of a diblock copolymer melt
    (American Physical Society, 2016) Beardsley, Thomas M.; Matsen, Mark W.
    The equivalent behavior among analogous block copolymer systems involving chemically distinct molecules or mathematically different models has long hinted at an underlying universality, but only recently has it been rigorously demonstrated by matching results from different simulations. The profound implication of universality is that simple coarse-grained models can be calibrated so as to provide quantitatively accurate predictions to experiment. Here, we provide the first compelling demonstration of this by simulating a polyisoprene-polylactide diblock copolymer melt using a previously calibrated lattice model. The simulation successfully predicts the peak in the disordered-state structure function, the position of the order-disorder transition and the latent heat of the transition in excellent quantitative agreement with experiment. This could mark a new era of precision in the field of block copolymer research.
  • Thumbnail Image
    Item type: Item ,
    Tangible World-in-Miniature Interaction in Virtual Reality
    (University of Waterloo, 2026-11-04) Kim, Junhyeok
    Virtual reality (VR) provides a spatial platform for working with miniature representations and tangible proxies across multiple scales and perspectives. However, current world-in-miniature (WiM)-based VR systems face persistent challenges in scale perception, interaction continuity, and the tradeoff between interaction precision and speed. These limitations collectively hinder intuitive multi-scale spatial interaction in VR. This dissertation investigates how tangible user interfaces (TUIs) with WiM-based VR can support interaction across multiple scales and views using tangible proxies. The research examines how users perceive and manipulate miniature representations, how visual–proprioceptive scale discrepancies affect interaction, and how such discrepancies can be managed through interaction design. To investigate this problem, I conducted a progressive, multi-stage research. First, I conducted an exploratory mixed-methods study examining three miniature design tools: online planner, paper-and-pencil sketching, and LEGO assembly. The study used observation, the Desirability Toolkit, the Flow Scale, and semi-structured interviews. Second, I conducted a controlled in-lab VR experiment examining visual–proprioceptive perceptual tolerance using tangible proxies paired with differently scaled virtual objects. Measures included interaction accuracy, task completion time, perceived scale estimation, and subjective discrepancy. Finally, I synthesized these empirical insights and prior literature into a multidimensional design framework, then instantiated one configuration of the framework as Laminar, a prototype interaction technique. This dissertation presents empirical results on how people use and experience three miniature-based interior design tools. It provides quantified evidence of users’ perceptual responses to visual-proprioceptive mismatch when manipulating tangible proxies against differently scaled virtual objects in VR. Finally, I present a design framework and instantiated interaction technique, Laminar, that integrates TUI and WiM within VR to support multi-scalar interaction. Together, this dissertation delivers evidence-informed interaction techniques and design frameworks for tangible WiM in VR, addressing collective challenges in scale perception, multi-scale manipulation, and multi-view workflow fragmentation. This work establishes theoretical and practical foundations for seamless, intuitive, and scalable multi-scalar spatial interaction in VR.
  • Thumbnail Image
    Item type: Item ,
    Biomechanics of Planar Collagenous Soft Tissues: Heterogeneities and Averaged Material Descriptions
    (University of Waterloo, 2026-11-03) Amiri, Arya
    The overarching goal of this thesis is to advance the understanding of biomechanical heterogeneities and averaged material description of planar collagenous soft biological tissues through experimental, constitutive, and computational approaches, while also refining and adapting existing methods to better suit the specific characteristics and complexities of soft tissues. Despite decades of research documenting the mechanical heterogeneity of soft tissues, existing frameworks have yet to be fully reconciled with the experimental evidence. As the first objective, reindeer back skin and velvet (antler skin) are mechanically tested using a biaxial tensile testing machine, providing a basis for demonstrating the significance of regional mechanical heterogeneity. Motivated to accurately represent each tissue region, the second objective evaluates the concept of averaged material descriptions, specifically ``response averaging" and ``parameter averaging", across various hyperelastic constitutive models, ranges of deformations, and test types, to accurately represent groups of specimens. These findings are further used in the third objective to study whether it is possible to accurately describe local spatial biomechanical heterogeneity with sufficient precision from only a few samples. Specifically, it is done by applying the averaging methods to both experimental stress-strain data and constitutive model outputs to accurately predict mechanically untested intermediate regions between adjacent mechanically characterized tissue regions. Having validated the predictive power of the averaged-material-description approach, the fourth objective of this thesis develops a framework for incorporating local spatial mechanical heterogeneity into finite element (FE) simulations of soft biological tissues. Specifically, motivated by functionally graded materials and using an averaging approach to describe local heterogeneity it proposes to model heterogeneous material constants expressed as continuous Fourier functions of spatial positions. Through case studies of aortic aneurysms and skin tissue, it is demonstrated that a continuous heterogeneous material definition leads to a different stress distribution compared with a homogeneous model and avoids stress concentrations associated with a segmental heterogeneous model. Hence, the integration of regional mechanical characterization and the proposed averaging techniques, combined with the developed method for implementing the mechanical heterogeneity into FEM simulations, explored across all objectives of this thesis, sets a new standard for studying soft tissue biomechanics.