21090 research outputs found
Sort by
Quantifying and Mitigating Uncertainty in Crash Risk Prediction for Road Safety Analysis
No full textRoad safety analysis is a cornerstone of traffic safety management programs like Vision Zero, which aim to eliminate fatalities and serious injuries on roadways. Central to road safety analysis is the ability to accurately predict crash risk; however, this task is challenged by significant uncertainty arising from the random nature of crashes (aleatoric uncertainty) and limitations in data and modeling (epistemic uncertainty). These uncertainties can lead to the misidentification of hazardous locations, resulting in false positives and negatives, and the inefficient allocation of limited safety resources.
While numerous statistical models exist for risk prediction, most traditional crash-based approaches provide simple point estimates, failing to formally quantify the inherent uncertainty in their predictions. Proactive conflict-based analysis has emerged as a promising alternative that avoids direct reliance on sparse crash data, but its application introduces new methodological challenges. The reliability of conflict-based predictions is not well understood, and key methodological choices, such as the duration of data collection and the selection of analytical thresholds for Extreme Value Theory (EVT) models, introduce significant, often unaddressed, uncertainty into the results.
To overcome these challenges, this thesis systematically develops and evaluates a framework to quantify, investigate, and reduce critical sources of uncertainty in road safety analysis. First, to quantify the impact of uncertainty on network screening, a frequentist approach is employed to establish a joint confidence region (CR) for hotspot rankings, moving beyond simple point estimates. This is achieved by first estimating the confidence interval (CI) of risk for each location using a hierarchical Full Bayesian (FB) model that considers both crash frequency and severity. Second, this research investigates a primary source of data uncertainty in conflict-based analysis by systematically assessing the relationship between sample size and prediction reliability using a unique, year-long LiDAR dataset and a Bayesian Peak-Over-Threshold (POT) EVT model. Third, to address methodological uncertainty in EVT, an automated and objective approach for threshold selection is developed and validated, comparing a Sequential Goodness-of-Fit Selection Method (SGFSM) with an Automatic L-moment Ratio Selection Method (ALRSM) to reduce analytical subjectivity.
The analysis demonstrates that explicitly accounting for uncertainty can lead to substantially different hotspot identifications, revealing that rankings based on point estimates alone may be unreliable. The sample size analysis reveals that the common practice of using short-term conflict data is inadequate for reliable collision predictions, a finding that challenges the validity of a significant portion of the existing literature on conflict-based safety analysis. Finally, the automated threshold selection approach, particularly the L-moment-based approach, proves to be a robust and objective method that improves the accuracy of crash risk estimation. Collectively, this research provides researchers and practitioners with an evidence-based methodology to understand, quantify, and mitigate key uncertainties in road safety analysis, fostering more reliable safety assessments and a more effective allocation of resources
Essays on Mobile Banking Adoption
No full textMobile banking (m-banking) has transformed financial services by enabling seamless access.
This dissertation examines psychological factors, mental health, device type, and m-banking adoption in Canada through three studies.
The first study conducts a systematic literature review on intrinsic factors in m-banking adoption, categorizing influences into psychological, personal, perceptive, and social dimensions.
Using text-mining and statistical techniques on 143 studies, it highlights overlooked trends and promotes diversified research approaches.
The second study analyzes mental health's association with m-banking adoption in Canada, integrating into it moderating effects of relationship satisfaction, smartphone dependency, and social media usage.
Using fixed-effect logistic regression on Canadian Internet Usage Survey data, it finds that better mental health correlates with lower adoption rates, while social media users and smartphone-dependent individuals show higher adoption tendencies.
The third study compares smartphone and smart wearable users in m-banking adoption.
Analyzing 18,000 survey records with logistic regression, it finds perceived trust is more crucial for smartphone users, while wearable users prioritize time-saving.
Demographics relate to m-banking adoption differently across devices, emphasizing the need for tailored financial strategies.
This dissertation bridges academic and industry perspectives, advancing technology adoption models and offering insights for financial institutions.
By integrating psychological and behavioral aspects, it lays a foundation for more user-friendly banking applications.
Addressing research gaps, it explores current adoption trends and the role of emotional states in shaping mobile banking behaviors, particularly within the Canadian context
The Dynamical States and Mass Accretion Histories of Galaxy Clusters in IllustrisTNG
No full textThe concordance cosmological model describes the history and large-scale structure of
the universe using a few key parameters. Two of these parameters, σ8 and Ωm, determine
the clustering of matter due to the growth of density fluctuations in the early universe.
Current constraints on these parameters measured from nearby large structures and from
the early universe are in statistical tension. Our ability to resolve this tension is limited
by the degeneracy between the parameters when measured from observations of nearby
structure. Since galaxy clusters are the most massive gravitationally bound objects in
the universe, their formation is sensitive to σ8 and Ωm. An improved understanding of
the formation histories of galaxy clusters can break the measurement degeneracy, thus
providing new insights into this tension in our cosmological model. Since the formation
time scales of galaxy clusters are unobservable, we must use the structure of clusters to
probe their formation histories.
In this thesis, we relate the observable structural properties of galaxy clusters to the
mass accretion histories of their surrounding dark matter halos in the IllustrisTNG cosmo-
logical simulations. Structure formation in the universe is hierarchical, so recently formed
galaxy clusters will have experienced recent mergers with other systems. We examine a set
of structural properties that are related to the dynamical states of clusters as indications
of recent mergers to relate the structures of clusters to their formation histories. Using the
cluster formation history information that is available in IllustrisTNG, we classify clusters
as dynamically relaxed or unrelaxed based on their structural properties and compare the
mean mass accretion histories of the resulting groups.
We establish in this work that the stellar mass asymmetry and magnitude gaps of
galaxy clusters are readily observable structural parameters that most effectively predict
the mass accretion histories of halos. By comparing the gravitational lensing profiles of
dynamically relaxed and unrelaxed clusters classified using different structural parameters,
we demonstrate that the stellar mass asymmetry most reliably distinguishes between halos
in different dynamical states with different density profiles. We also show that line of
sight galaxy projection does not significantly affect IllustrisTNG cluster samples and that
differences between 3D-identified clusters and optically selected clusters can be accounted
for with accurate cluster mass estimates. However, we find that the density profiles traced
by the weak gravitational lensing around relaxed and unrelaxed clusters in IllustrisTNG
simulations and DLIS x UNIONS observations are discrepant. The structural differences
between the simulated and observed galaxy clusters will be further explored in future work
to better relate this work’s findings to real astrophysical systems.
Overall, we find through cosmological simulations that the structural properties of
galaxy clusters can be used to effectively trace their mass accretion histories. The findings
of this thesis establish which observable properties of clusters can be targeted in both
observations and simulations to grant us insight into the formation histories of the largest
structures in the universe
Intelligent Multi-Robot Autonomy with Connected AMRs and Manipulators for SMART Factory(s)
No full textThe transition toward SMART factories demands robotics systems that go beyond conventional automation to enable intelligent, autonomous, and scalable operations. This thesis presents a unified multi-robot autonomy framework that integrates distributed 3D mapping, 4D radar-based perception, 5G wireless communication, and high-DoF collaborative manipulation to address the challenges of modern industrial environments. The proposed system comprises two novel synergistic verticals: Connected Robotics Architecture for Distributed SLAM Mapping (CRADMap), a distributed volumetric mapping architecture for multi-robot systems using Autonomous Mobile Robots (AMRs), and Radar Antenna Pattern Acquisition through Automated Collaborative Robotics (RAPTAR), a radiation scanning and acquisition platform for radar antenna characterization using collaborative manipulators for enhancing HRI (Human Robot Interaction). CRADMap enables novel volumetric SLAM algorithm development, real-time 3D reconstruction by offloading dense RGB-D and radar data from AMRs to a centralized backend via 5G, where data is fused using COVINS for globally consistent map generation. The novel automation of 4D mmWave radar enhances perception in occluded or cluttered spaces, enabling inspection beyond line-of-sight. RAPTAR automates the traditionally manual process of radiation pattern testing using a 7-DoF torque-controlled cobot equipped with a custom end-effector, executing smooth, azimuth-polar constrained trajectories synchronized with RF data acquisition without the need for anechoic chambers. Together, these systems demonstrate a deployable ROS2 Humble, C++-based software stack, developed and validated through real-world experiments. Key novel contributions include: (i) distributed SLAM for multi-robots (AMRs), (ii) radar-augmented volumetric perception, (iii) Edge compute-enabled data pipelines using 5G, and (iv) automated high-resolution robotic manipulation for radiation measurement. This thesis establishes a practical blueprint for next-generation SMART factories, agents operate collaboratively to perceive, decide, and act autonomously and safely in dynamic, and data-driven industrial ecosystems
Acceleration of Integer Transformer Models Via Structured Resource Management Using FPGAs
No full textThe widespread adoption of Large Language Models (LLMs) in various applications has pushed the demand for efficient hardware acceleration beyond the capabilities of traditional platforms. Due to their highly parallel architecture and ease of deployment, Field Programmable Gate Arrays (FPGAs) are widely used to accelerate LLMs. However, the FPGAs’ limited on-chip memory resources are still too limited to accommodate the trained models. While existing FPGA-based solutions have demonstrated promising throughput and energy efficiency, they often rely on abundant fabric resources, assume high-bandwidth devices that are not suitable for deployment at the edge, or employ highly customized acceleration architectures that are not scalable with the advancements of the LLMs architectures. This thesis addresses these challenges by proposing a novel on-chip resources manager architecture for integer encoder-based transformer inference, with a focus on Bidirectional Encoder Representations from Transformers (BERT) models. We target resource-constrained FPGAs with limited memory bandwidth. We show that, through structured operation scheduling and resource-sharing, significant performance improvements can be achieved. The proposed resource-shared infrastructure is also designed to be modular, allowing newly introduced computation blocks to be easily integrated into the accelerator without requiring major modifications or incurring additional off-chip data movement.
Demonstrated on a fully quantized integer-only variant of the BERT model as a representative workload, the proposed system achieves 2.32x latency improvement over the baseline custom accelerator, 1.17x over Jetson Orin Nano GPU, and at least 23.63x over CPU. The design is validated on two FPGAs: the PYNQ-Z1 as a low-end proof-of-concept and the KV260 as a mid-range deployment target
Evaluating a priori and data-driven weighting of the Healthy Eating Food Index-2019 for assessing diet quality and gastrointestinal and aerodigestive cancer risk in Canadian adults
No full textBackground: Diet is a modifiable exposure implicated in gastrointestinal and aerodigestive cancers. Because foods are consucmed in combination, diet quality indices are used to summarize overall dietary patterns. The Healthy Eating Food Index-2019 (HEFI-2019) measures adherence to Canada’s Food Guide 2019, and its component scores are nearly equally weighted, reflecting the importance of all foods in a healthful dietary pattern. Its discriminatory capacity for measuring diet-disease associations, and the influence of the weighting schema of the index, remains uncertain.
Objective: To assess whether associations between diet quality and gastrointestinal and aerodigestive cancer risk differ among adults in Canada based on the a priori Healthy Eating Food Index-2019 (HEFI-2019) versus a novel modified version with components reweighted using a data-driven approach.
Methods: A prospective cohort analysis was conducted using the Canadian Community Health Survey 2004 Nutrition (CCHS 2004) linked with the Canadian Cancer Registry (CCR) through 2016. After exclusions, 10,530 adults were included, representing approximately 23.5 million Canadians. Diet was assessed using interviewer-administered 24-hour recalls. HEFI-2019 total scores were computed using standard weights and using data-driven weights derived from ridge-penalized Cox models in 10 iterations of 80/20 training–test splits with cross-validated penalty selection. Weighted Cox proportional hazards models, adjusted for age, sex, education, income, marital status, smoking status, body mass index, and alcohol consumption, estimated associations with incident gastrointestinal and aerodigestive cancers (ICD-9 140–149, 150–159, 160–161). Discrimination was assessed with Harrell’s C-index.
Results: The data-driven approach altered component weights substantially (e.g., protein foods increased from 5 to 16.4; vegetables and fruits decreased from 20 to 3.73). No associations with cancer risk were observed for either the a priori (adjusted HR per unit increase 1.01; 95% CI: 0.99, 1.04) or reweighted HEFI-2019 scores (adjusted HR: 1.00; 95% CI: 0.98, 1.02). Model discrimination was similar (Harrell’s C-index: 0.81 [95% CI: 0.77, 0.85] for a priori; 0.87 [95% CI: 0.80, 0.93] for reweighted).
Discussion: Neither the a priori nor reweighted HEFI-2019 was associated with gastrointestinal and aerodigestive cancer risk. Data-driven reweighting did not meaningfully improve associations or discriminatory capacity. These findings suggest challenges in using diet quality indices for complex diet-disease relationships and highlight the need for further research on index construction and application in cancer epidemiology
Improving Automated Lung Ultrasound Interpretation with Self-Supervised Learning
No full textLung ultrasound (LU) is an increasingly important point-of-care examination in acute healthcare settings. In addition to exhibiting comparable accuracy to conventional imaging modalities such as radiography or computed tomography, ultrasound provides safety from radiation and enhanced portability at a reduced cost. Despite its purported benefits, LU has yet to be widely adopted due to a lack of trained experts and education programs. In response, machine learning algorithms have surfaced to alleviate the skills gap by providing automated interpretation. However, the training of machine learning algorithms requires vast amounts of manually annotated examples - that is, images labelled by an expert for whether or not they exhibit particular findings. Given that there are few experts qualified to provide quality annotations, there is a need to explore alternative machine learning techniques for this unique problem.
Self-supervised learning (SSL) has emerged as a class of methods to train machine learning algorithms to extract salient features from data without relying on labels. These so-called “pretrained” algorithms constitute better starting points when unlabelled data is abundant but labelled examples are scarce or challenging to acquire, as is the case for LU. Therefore, the purpose of this thesis is to propose and evaluate novel techniques tailored to lung ultrasound that enhance performance on core interpretation tasks, such as detecting lung sliding and pleural effusion.
The thesis consists of four studies. The first study explores the efficacy of SSL techniques in brightness mode LU interpretation tasks and proposes a multi-task framework that reuses a single pretrained algorithm to efficiently interpret LU images. The results demonstrate that SSL improves performance on multiple classification tasks. In addition, experiments show that pretrained algorithms amplify generalizability to external healthcare centres. The second study introduces SSL methods to motion mode ultrasound and proposes data augmentation techniques specific to motion mode. We observe that pretrained algorithms achieve the greatest performance on local and external test data for the challenging task of lung sliding classification. The third study proposes a LU-specific technique for SSL that involves sampling pairs of images from the same ultrasound video that are temporally or spatially proximal to each other, based on the intuition that such pairs of images share similar content. The results indicate that, with appropriate parameter assignments, this sampling strategy improves performance of pretrained algorithms on multiple tasks. Lastly, the fourth study proposes ultrasound-specific data augmentation and image preprocessing methods for SSL. The results underscore the value of ultrasound-specific image preprocessing in SSL. A comprehensive evaluation finds that ultrasound-specific data augmentation yields the best performance on a diagnostic task, and that techniques based on cropping attain top performance on object classification tasks.
Overall, the findings of this thesis demonstrate that SSL improves the performance of machine learning algorithms for LU, especially on LU images originating from external healthcare centres. Novel SSL methods for LU are established as a key ingredient for producing algorithms that are effective for multiple LU interpretation tasks
Figuring Forgiveness: Dramatistic Aspects of Forgiveness in an Anabaptist Context
No full textThis dissertation is a knowledge-translation project of artistic creation: the composition and performance of three sermons which bring Kenneth Burke’s rhetorical method to the topic of forgiveness in an Anabaptist context. It faces the special challenge of explaining from the cognitive perspective of a Burkean rhetoric of motives why Anabaptists – particularly Amish and Mennonites – forgive. And it faces the challenge of explaining that to an Anabaptist congregation. I ask, “What is involved when we say we are forgiving, and why are we doing it?” Where other treatments of the topic have foregrounded sociological or historical perspectives, this project illuminates the suasive and formative qualities of forgiveness as a distinctly rhetorical act, and comes at the topic from a perspective situated as both rhetorician and pastor within the Anabaptist tradition. The sermons not only function to communicate the analytical and substantiating power of Burke’s rhetorical method, but also enact that power in homiletic performance.
As instances of knowledge mobilization, the sermons translate and apply the theoretical valence of Burke’s dramatism to the practical and contextualized task of preaching. In particular, the sermons mobilize Burke’s concept of identification and his theories of form to illustrate the rhetorical dimensions of forgiveness in divine, social, and personal domains. As creative pieces within an expository framework reflecting the homiletical vein of the rhetorical tradition, the sermons also channel Burke’s voice as a literary critic and explore Anabaptist texts such as confessions of faith, martyrologies, hymnals, and devotional books as “equipment for living,” while at the same time directly offering Anabaptist literary equipment themselves in the performance of the sermons.
The first sermon explores forgiveness as an act nested within a scene of divine drama, framed within an exposition of Romans 5:1-10. The second sermon prioritizes aspects of the Agent:Act ratio within an exposition of Matthew 18:21-35 to explore how interpersonal identification shapes attitudes toward receiving and extending forgiveness. The third sermon prioritizes the Act:Agent ratio within an exposition of the Lord’s Prayer from Matthew 6, exploring the formative relationships between language, devotional practices and attitudes of forgiveness. These three sermons are framed by introductory and concluding chapters which provide the theoretical context and offer a scholarly and expanded consideration of how Burkean rhetorical theories relate to forgiveness in Anabaptist practice and literature. The texts of the sermons are provided in both bare and annotated forms; the annotated versions provide additional scholarly analysis, with footnotes addressing performative, perlocutionary elements while the endnotes address broader analytical and critical features. The opening and closing chapters theorize the project, framing its autoethnographic features and situating it within broader questions of my own identity as a Mennonite scholar, pastor, and preacher. Ultimately, this dissertation argues, through both the literary performances and the scholarly apparatus, that a full comprehension of forgiveness in an Anabaptist context means understanding its broader rhetorical dimensions, and that the application of a Burkean rhetoric of motives provides a more rounded appreciation of the symbolic forces that both form and are informed by Anabaptist values and beliefs about forgiveness
Investigating the Role of the Femoral Tunnel Position on Knee Biomechanics in ACL Reconstruction
No full textThe femoral tunnel position for the graft plays a critical role in the success of anterior cruciate ligament (ACL) reconstruction surgery. Malposition of the femoral tunnel is one of the top reasons for graft failure in this procedure. Despite the potential benefits of surgical navigation systems for ACL reconstruction to aid in more accurate femoral tunnel positioning, there has been no significant adoption or evidence of improved clinical outcomes with navigated ACL reconstruction procedures. However, this may stem from a lack of understanding of how the femoral tunnel position in ACL reconstruction alters the biomechanics of the knee joint relative to the native knee. To address this knowledge gap, the goals of this work were to (1) investigate the role of the femoral tunnel position in its influence on knee biomechanics, and (2) design and develop a proof-of-concept surgical navigation system to aid in selection and placement of the femoral tunnel. For the first goal, an in vitro study investigated the influence of five different femoral tunnel positions on knee kinematics, force carried by the graft, and on graft length change over flexion. Varying femoral tunnel insertion led to statistically significant kinematic differences in anterior tibial translation range at 90 degrees of knee flexion. As well, different femoral tunnel positions led to significant differences in the force carried by the graft compared to the estimated force carried by the intact ACL at all knee flexion angles. Graft length change over knee range of motion showed that an anteriorly placed femoral tunnel is more isometric while a posteriorly placed graft had the greatest change in length. For the second goal of this work, a surgical navigation system workflow was developed based on these insights to allow for the calculation of graft length changes intraoperatively over different knee flexion angles. A proof-of-concept system with this feature was developed in 3D Slicer using the SlicerIGT extension. This work enhances understanding of how the femoral tunnel position affects knee biomechanics and how surgical navigation systems can be designed to provide surgeons with methods to better identify and position the femoral tunnel for ACL reconstruction surgery
Towards a Holistic Understanding of the Drivers of Media Multitasking
No full textMedia multitasking, defined as simultaneously engaging with multiple tasks when at least one of the tasks involves media, is a highly prevalent behaviour known to negatively impact performance across a variety of contexts. It is therefore crucial to understand what motivates individuals to engage in this behaviour. The studies presented in this thesis examined participants’ decisions to engage in media multitasking during sustained attention tasks, along with the predictors of these decisions, to provide a more nuanced understanding of the drivers of media multitasking. Chapter 1 discusses the current media multitasking literature, highlighting a need for further research investigating the variables that underly changes in this behaviour over time, as well as research investigating the combined role of individual differences and contextual factors in media multitasking. Chapters 2 and 3 explored whether variables relevant to the perceived opportunity costs of completing a task, namely boredom and motivation, are associated with changes in media multitasking over time. Chapter 2 provided evidence that rising opportunity costs, signaled by increases in boredom over time, may drive temporal increases in media multitasking. Across two studies, Chapter 3 demonstrated that increasing motivation attenuates temporal increases in media multitasking. Chapters 4 and 5 examined the joint role of individual differences and contextual factors in predicting media multitasking. Chapter 4 specifically explored the interactions between these two classes of factors, revealing that relations between in-study media multitasking and individual differences in attention-related and self-regulatory traits, as well as real-world media multitasking tendencies varied based on two key contextual variables: whether participants were completing an easy or challenging task and the order in which they completed these tasks. Chapter 5 tested the assumption that, if media multitasking arises from a combination of individual differences and contextual factors, then when context is held constant over time, patterns of media multitasking should remain consistent. Moreover, this consistency should largely be driven by individual differences. Media multitasking was assessed in the same laboratory context on two separate occasions and was found to be consistent across sessions. Additionally, several individual difference factors interacted with session when predicting media multitasking. For the most part, relations involving these factors varied in strength, but not direction across sessions. The final chapter (Chapter 6) summarizes the work presented in this thesis, contextualizes it within the current literature, and emphasizes the need for further work investigating how factors within various layers of influence interact to give rise to media multitasking. This chapter also proposes a novel framework for considering and guiding future research on these interactions