21793 research outputs found
Sort by
Explicit and Implicit CSR: An Exploration of the Canadian Energy Sector
No full textCorporate Social Responsibility (CSR) has evolved into a fundamental component of corporate identity and stakeholder engagement since the 1950s. Research has focused on cross-national and industry comparisons without much attention to potential subnational and intra-industry differences. This thesis examines how CSR evolves over time through the interaction of regional institutional contexts and global societal expectations.
The longitudinal comparative case study design explores CSR narratives of two Canadian energy firms, Hydro-Québec in Québec and Suncor in Alberta between 2010 and 2024. Annual reports and press articles are used for qualitative content analysis to assess firm self-presentation and media framing. This thesis builds on Matten and Moon’s (2008, 2020) frameworks on explicit and implicit CSR including institutional theory and National Business Systems (NBS).
Findings show that firms in the same industry and country, regardless of the nature of CSR will exhibit explicitization over time. However, this process operates through divergence and convergence. Across dispersed territories, institutional proximity outweighs industry affiliation. Institutional proximity will create different CSR foci intra-industry because of ownership and regional influences. However, CSR discourse will show more convergence on global themes like clean energy, emissions reduction, and Indigenous reconciliation. News outlets often reflect these narratives, while exposing differences between self-presentation and perception.
Through the longitudinal, subnational, and intra-industry lens, this thesis presents the ever-changing nature of CSR communication affected by institutional logics, ownership, and regional context. Contributions to the field include emphasis on subnational differences, multi-source analysis, changes in CSR, and the existence of intra-industry variation
On Carbon Dot Based Drug Delivery Systems: Mechanochemical Formation of Surface Imine Bonds
No full textEfficient delivery of hydrophobic drugs remains a significant challenge, often hindered by poor distribution within aqueous bloodstreams. Current medical literature shows that over 40% of all drugs in clinical use are hydrophobic, and 90% or more of newly developed drugs are described as Class II (high permeability, low solubility), or Class IV (low permeability, low solubility). Carbon dots (CDs) offer a promising solution as drug delivery vehicles due to their aqueous dispersibility, generally low cytotoxicity and ease of surface functionalization. Traditional conjugation methods widely rely on covalent bonds such as amide bonds, which are robust but difficult to cleave, hindering drug release. Furthermore, their formation often requires expensive and hazardous coupling agents such as 1-Ethyl-3-(3-dimethylaminoporpyl)carbodiimide. Herein and to the best of our knowledge, we report for the first time, the solid-state mechanochemical formation of imine bonds to conjugate a hydrophobic model drug to the surfaces of CDs. Initial efforts focused on solution-based conjugation before transitioning to mechanochemistry, which was ultimately successful. 1H-NMR analysis confirmed successful formation of imine bond linkages through the conversion of an aldehyde into an imine functional group without the need for a solvent, or any additional reagents. Additionally, release of the model drug was achieved reaching a plateau after 24 hours. A preliminary resazurin cytotoxicity assay demonstrated the low cytotoxicity of the CDs before and after altering the surfaces with imine bonds. Our findings highlight a straightforward and sustainable method for CD functionalization, paving the way for alternative covalent linkages and greener conjugation strategies
High-Frequency Forecasting of Bitcoin Volatility: Evaluating HAR Models with Realised Semivariance and Jump Components
No full textBitcoin’s continuous trading, speculative nature, and high volatility create distinctive challenges for risk management and forecasting. This thesis examines how high-frequency realised volatility (RV) measures and Heterogeneous Autoregressive (HAR) models capture Bitcoin’s volatility dynamics and improve forecast accuracy. Adapting RV methods from equity markets, the analysis adds downside semivariance to address asymmetric negative returns and jump variation to capture price movements.
Using minute-level Bitcoin prices from, I compute RV from 5-minute returns and estimate four HAR variants—baseline HAR, HAR-RS, HAR-J, and HAR-RS-J. Models are re-estimated in a rolling window, and forecasts are evaluated with RMSE, MAE, and QLIKE. Robustness checks test stability under different data granularities and market regimes. A GARCH(1,1) benchmark provides a parametric comparison, with HAR variants outperforming it at short horizons, while GARCH exceeds performance at longer horizons.
Results show that HAR-type models capture Bitcoin’s long memory, volatility clustering, and asymmetry effectively. HAR-J delivers the most accurate day-ahead, while HAR-RS leads at weekly and monthly horizons due to persistent downside risk. At quarterly horizons, forecast accuracy converges across models as high-frequency information loses relevance.
This study extends RV–HAR modelling to cryptocurrency markets, revealing shorter volatility persistence and greater jump contributions than in equities. It identifies downside semivariance and continuous variation as robust predictors across different market conditions and offers horizon-specific tools—HAR-J for short-term risk management and HAR-RS for medium-term volatility planning
Deregulation and the $10 Billion Threshold: Bank Behavior, M&A Activity, and Market Reactions Post-EGRRCPA
No full textThis thesis examines how the Economic Growth, Regulatory Relief, and Consumer Protection
Act (EGRRCPA) affected U.S. banks near the 10 billion threshold through mergers and acquisitions
(M&A), suggesting a strategic shift enabled by the rollback of Dodd-Frank provisions.
The findings demonstrate that deregulation spurred growth and consolidation, particularly among
smaller banks, highlighting how rigid regulatory thresholds can distort bank behaviour. The study
contributes to the literature by documenting the real effects of deregulatory policy on financial
institutions’ strategies and market performance.
ii
Utilizing Autonomous Vehicles to Reduce Truck Turn Time in Ports with Application for Port of Montréal
No full textThis thesis develops a Discrete Event Simulation (DES) model to evaluate strategies for reducing truck turn time (TTT) and enhancing operational efficiency at the Port of Montreal's Viau Terminal. The model analyzes the complex landside operations, including gate processes, internal movements, staging, and yard handling, differentiating between Human-Driven Vehicles (HDVs) and Autonomous Vehicles (AVs) based on their distinct behavioral and efficiency attributes. The study aims to provide insights into the potential of AV integration and demand management strategies in mitigating port congestion. DES, with integrated agent-based logic, is employed to simulate four distinct scenarios: a baseline representing current operations, a scenario implementing a Truck Appointment System (TAS) only, a scenario with partial AV integration (35% AVs) under shared resources, and a final scenario with AVs benefiting from dedicated staging areas and partitioned yard cranes. The model's conceptualization is informed by real-world data from port cameras and official reports, and its credibility is established through rigorous verification and validation against observed TTT metrics. The simulation findings reveal that the baseline scenario exhibits an average TTT of 88.2 minutes, characterized by significant internal congestion. The introduction of a TAS reduces TTT to 78.37 minutes. Partial AV integration (Scenario 3) further decreases the overall TTT to 55.91 minutes, with AVs achieving a TTT of 45.33 minutes. The scenario with dedicated AV staging and cranes (Scenario 4) results in the lowest AV TTT of 32.86 minutes; however, the overall system TTT unexpectedly increases to 57.13 minutes, as HDV TTT rises to 70.20 minutes. The study concludes that a multi-faceted approach involving demand management, vehicle technology, and strategic investment in infrastructure has a key function in maximizing port efficiency. Quantitative evidence and actionable recommendations are offered in this research to port authorities, with an emphasis on the necessity of nuanced resource allocation plans in the evolution towards automated port logistics
From Top-Down Interventions to Community-Based Practices: Gender and Social Transformations in Rural Iran
No full textThis dissertation investigates the transformation of gendered land relations in rural Iran over the past seven decades, with a particular focus on the Caspian provinces of Gilan and Mazandaran. It examines how state-led agrarian reforms, especially those implemented during the White Revolution (1962-1971), alongside more recent grassroots initiatives in the 2010s and early 2020s, have shaped women’s access to land, agricultural labour, and socio-economic visibility. Bridging historical analysis, ethnographic fieldwork, and digital ethnography, the study draws on both archival data and intergenerational family narratives to trace the ways in which rural women have been systematically marginalized in formal land policy while also participating in and imagining more equitable, ecologically grounded alternatives. The research is grounded in ecofeminist, postcolonial feminist, and rural sociological frameworks. It critiques development paradigms that overlook the significance of subsistence labour and care work, and it exposes the gendered biases embedded in Iranian land entitlement and agrarian policy. The study also explores the contemporary back-to-the-land movement in northern Iran, largely led by urban-educated youth, as a site of both opportunity and tension. These mostly family-run ecological farms are examined for their potential to revalue care labour and promote more equitable gender and class dynamics, particularly as many turn to agritourism to support financial sustainability. While enhancing the financial sustainability of these small-scale economies, this strategy often entails the commodification of traditional rural life, including gendered roles and expectations rooted in cultural memory. Overall, this dissertation provides a contextualized understanding of how rural women navigate systemic constraints by situating rural transformations in Iran within broader global debates on land, gender, and environmental justice, as well as the exercise of agency and the articulation of alternative futures through everyday practices and shifting socio-political imaginaries
Involvement of Corticotropin-Releasing Factor Signaling in Food Deprivation Stress-Induced Heroin Seeking Following Punishment-Imposed Abstinence in Male Rats
No full textRelapse remains a major challenge in treating opioid use disorder, often triggered by
stress-related factors that reinstate drug-seeking after abstinence. Corticotropin-releasing factor (CRF), a neuropeptide central to the stress response, has been implicated in stress-induced relapse, particularly via the mesolimbic dopamine system. However, the neural sources of relapse-driving CRF remain unclear. This thesis investigates CRF signaling in stress-induced heroin seeking and examines the paraventricular nucleus of the hypothalamus (PVN), a CRF expressing region, as a potential driver.
In the first experiment, male rats were trained to self-administer heroin using a seek-take chain schedule modeling human drug use. After stable intake, rats underwent punishment-imposed abstinence, where footshocks were probabilistically paired with drug-seeking responses, leading to suppression. During relapse testing, each rat was evaluated under two conditions: sated and food deprived. Heroin-seeking behavior was significantly higher under food deprivation than when sated. This increase was blocked by intracerebroventricular administration of a CRF receptor antagonist, supporting CRF’s role in stress-induced relapse and validating the seek-take punishment model.
In the second experiment, the PVN was targeted to explore its contribution to relapse.
Rats underwent the same protocol and received either a ligand to chemogenetically inhibit PVN neurons or vehicle before relapse testing. PVN inhibition modestly reduced heroin seeking under food deprivation but did not reach strong statistical certainty. These results suggest the PVN may contribute to CRF-driven relapse, though further studies are needed.
Together, these findings highlight CRF’s role in stress-induced relapse and suggest involvement of multiple CRF-releasing regions
Curating Fear: Reframing News for Mental Health
No full textIn today’s media landscape, the overwhelming volume of news consumption can lead to significant mental health impacts, including stress, information overload, and disengagement. This often results in news avoidance or a turn toward disinformation as a coping mechanism. This thesis advocates for a paradigm shift in how news is delivered and consumed, exploring the potential use of artificial intelligence as a tool for intervention and the prioritizing of mental well-being. This research-creation project presents a news chatbot, which functions both as a conceptual prototype and a critical site for examining AI's evolving role in journalism, to help users balance staying informed with preserving their mental health. The chatbot serves as a conceptual demonstration of how digital innovation, particularly within journalism, can integrate care, reflection, and emotional awareness into the act of news consumption, potentially reengaging otherwise alienated people. Rather than accept harm as the cost of being informed, this project imagines how we might relate to information differently and begin to reclaim the humanity that's been lost in the process
Greedy sparse recovery algorithms: from weighted generalizations to deep unrolling
No full textSparse recovery is a clear manifestation of Occam’s razor in the mathematics of data thanks to its ability to favor the simplest explanation. A prominent example is the reconstruction of a sparse
vector (i.e., one with mostly zero or negligible coefficients) from linear measurements, possibly corrupted by noise. This problem arises in numerous applications across various domains, including medical imaging and high-dimensional function approximation. Greedy sparse recovery algorithms approach this problem by recursively solving local optimization problems and have proven to be efficient alternatives to convex methods.
In this dissertation, we first develop weighted generalizations of Orthogonal Matching Pursuit (OMP)–one of the most popular greedy sparse recovery algorithms–based on two distinct weighting strategies aimed at incorporating a priori knowledge about the signal structure. These generalizations feature explicit and theoretically justified greedy index selection rules. The first strategy establishes a novel connection between greedy sparse recovery and convex relaxation methods,
particularly the LASSO family, resulting in new OMP-based algorithms suited to a wide range of loss functions. Numerical results show that these greedy variants inherit key traits from their ancestor convex decoders. For the second strategy, we provide optimal recovery guarantees for rescaling-based weighted OMP under the weighted Restricted Isometry Property (wRIP), extending the classical RIP-based analysis to the weighted setting.
The second part of the thesis addresses the non-differentiability of greedy algorithms caused by the (arg)sorting operator by employing "softsorting", enabling differentiable, neural-networkcompatible versions of these algorithms. We provide rigorous theoretical analysis and numerically demonstrate that these "soft" algorithms reliably approximate their original counterparts. We then link our weighted greedy solvers to neural architectures, by embedding their iterations into neural networks using the “algorithm unrolling” paradigm, treating weights as meaningful trainable parameters. This approach not only advances the development of data-driven methods for sparse
recovery, but also marks a significant step toward building safer and more trustworthy neural networks. Finally, we extend this framework to deterministic compressed sensing and to learning additional parameters beyond weights
Optimizing Protocols for Combining Imaging Mass Spectrometry (IMS) and Optical Imaging of Traditionally Histologically Stained Tissues: Advancements in Single-Cell Analysis Using IMS.
No full textBiomolecular changes linked to disease can be studied by integrating Imaging Mass Spectrometry (IMS) with histopathology. However, co-registering IMS with optical images of stained tissues is challenging due to sample preparation constraints and resolution discrepancies, particularly as IMS advances toward single-cell resolution. This work evaluates workflows for multimodal tissue analysis (sequential vs. consecutive) by combining IMS with histological imaging and incorporating laser-etched indium tin oxide (ITO) slides to improve image registration at the cellular level.
In this study, coronal mouse brain tissues were analyzed using cluster ion beam secondary ion mass spectrometry (SIMS) and/or matrix-assisted laser desorption/ionization (MALDI) mass spectrometry. For high-resolution MALDI imaging, 1,5-diaminonaphthalene matrix was sublimated on the tissue sections. Traditionally, serial tissue sections have been used for hematoxylin and eosin (H&E) staining and IMS image co-registration. Here, we examine the benefits and caveats of staining the same tissue section post-IMS analysis.
The 35 µm spatial resolution of our TOF-MALDI instrument exceeds the average diameter of a mouse brain cell, limiting single-cell multimodal IMS analysis. To overcome segmentation challenges, we employed Cell Segmentation Globally Optimized (CSGO), an open-source deep learning model specifically optimized for histological images, which enables accurate and automated whole-cell segmentation from optical microscopy of H&E-stained tissue images. While consecutive sections align overall tissue structure, they fail at cellular precision due to misalignment. By combining deep learning-based segmentation with same-section multimodal co-registration using laser-etched fiducial markers, we achieved improved spatial alignment for high-resolution molecular mapping, advancing disease characterization and biomarker discovery