21793 research outputs found
Sort by
Quadratic convolutional and physics-informed neural networks with applications to system theory
No full textMotivated by a growing need for explainable artificial intelligence, this thesis explores quadratic neural networks (QNNs) as a solution to many issues that limit the application of neural networks to safety-critical systems. Specifically, this thesis shows that the training of both a two-layer convolutional quadratic neural network (CQNN) and a two-layer quadratic physics-informed neural network (QPINN), with no regularization and constraints added on their activation function parameters and the norm of their weights, can be formulated as a least squares problem. Using this method, an analytic expression for the globally optimal weights is obtained, alongside a quadratic input-output mapping for the network. These properties make the network a viable tool in system theory by enabling formal analysis. Furthermore, compared to backpropagation-trained networks, the least squares training significantly reduces both training time and the number of hyperparameters that the designer must select. Additionally, it is proven that two-layer QNNs become universal approximators when a monomial lifted input of sufficiently high degree is used. Finally, ensemble QNNs (EQNNs) are proposed to train multiple QNNs across the training space to increase the representational capacity of QNNs
Cyber Security in Event-Triggered Cyber-Physical Systems: A System Theoretic Approach
No full textCyber-physical systems (CPS) are the backbone of critical infrastructures such as power grids, transportation networks, and water treatment facilities. While CPS offer numerous benefits, they face significant cyber-security challenges due to their large cyber surface, which allows adversaries to launch cyber-attacks. These attacks can be stealthy, making systems unstable while remaining undetected. Key challenges in CPS security include cyber-attack detection and isolation, resilient control design against adversaries, and privacy preservation. Due to bandwidth limitation
challenges, designing CPS with event-triggered communication—where components transmit information only when necessary—can reduce network congestion and optimize communication bandwidth. This thesis addresses several cyber-security challenges in CPS with event-triggered communication. The first part focuses on cyber-attack detection and isolation. Detection mechanisms are developed to identify attacks such as covert attacks and zero dynamics attacks. By leveraging event-triggered communication, we propose mechanisms capable of detecting highly resourced attacks even when adversaries have full knowledge of the detection system. The methodology also includes isolation mechanisms to isolate under-attacked communication channels.
The second part studies resilient control design in event-triggered CPS. A resilient control protocol is developed against covert attacks using auxiliary systems and observers, ensuring the system state remains Uniformly Ultimately Bounded (UUB). For zero dynamics attacks, a bank of auxiliary systems estimates the attack signal, which is then incorporated into a resilient control protocol, ensuring UUB under this stealthy cyber-attack
as well.
The third part investigates a Multi-Agent System (MAS) framework, where we address event-triggered resilient consensus control under False Data Injection (FDI) attacks among heterogeneous agents. By using a node decomposition methodology, the proposed control protocol ensures both resilience to cyber-attacks and preservation of agents’ privacy.
Overall, this thesis considers both Cybersecurity challenges and bandwidth imitations by designing robust detection, isolation, and resilient control protocols against various cyber-attacks while maintaining efficient network operation through considering event-triggered communication
Individual difference characteristics and contextual factors affecting educational attainment
No full textThe studies in this dissertation examine cognitive and non-cognitive predictors of postsecondary educational attainment. While prior research has documented links between core cognitive abilities (e.g., processing speed, attention, fluid intelligence) and academic success, less is known about the mechanisms translating these abilities into outcomes. It also remains unclear how contextual disruptions, such as the COVID-19 pandemic impacted students’ psychosocial and academic functioning.
The first study investigated the role of learning strategies, along with willingness to engage in effortful cognitive activity (Need for Cognition; NFC), as potential intermediaries between basic cognitive abilities and academic outcomes. Results showed that while standard cognitive measures did not directly predict academic performance, both NFC and model-based (goal-directed) learning strategies were significant positive predictors. Further analyses indicated that fluid intelligence and attention positively predicted NFC and model-based learning, suggesting that these abilities may facilitate the development of motivational and strategic traits that, in turn, promote academic success. These findings emphasize the importance of motivation and strategy use, even when direct associations with basic cognitive abilities are lacking.
The second study complements the first by examining the broader socio-environmental challenges posed by the COVID-19 pandemic on Canadian university students, with a focus on understanding the impact of the pandemic on students’ mental health, social networks, SES, and educational attainment. Using longitudinal data collected before and during the pandemic, results revealed that while GPA slightly improved, psychosocial well-being deteriorated. Increases in substance use, smaller social networks, and reduced well-being were observed. Cross-sectional analyses further showed that greater substance use during the pandemic predicted poorer GPA, and students with pre-existing psychiatric conditions were particularly vulnerable to increased substance use. These findings suggest that students with mental health vulnerabilities may be disproportionately affected by crises, underscoring the need to address maladaptive coping to support academic success.
Together, these studies highlight both individual (e.g., cognition, motivation, learning strategies) and contextual influences (e.g., pandemic disruptions) as important predictors of academic attainment. By considering internal and external factors, this dissertation provides a more comprehensive understanding of the multifaceted determinants of educational success, informing both theory and practice for optimizing university student outcomes
Accurate Abstract Syntax Tree Differencing: Language-Aware Design, Benchmarking, and Empirical Assessment
No full textSoftware undergoes constant changes to support new requirements, address bugs, enhance performance, and ensure maintainability. As a result, developers spend a large portion of their workday understanding and reviewing code changes. Abstract Syntax Tree (AST) diff tools were developed to overcome the limitations of line-based diff tools, which are still the default for most developers. Despite their advantages in capturing structural changes, existing AST diff tools suffer from serious limitations, such as lacking multi-mapping support, matching semantically incompatible nodes, ignoring language-specific clues, lacking refactoring awareness, and offering no commit-level diff support.
To address these issues, we propose a novel AST diff tool based on RefactoringMiner that resolves all aforementioned limitations. We improve statement mapping accuracy and introduce an algorithm that produces commit-level AST diffs using refactoring instances and matched program elements. Our evaluation demonstrates significant improvements in both precision and recall, while maintaining competitive execution times.
To facilitate objective and reproducible assessment of diff quality, we introduce a benchmarking framework that measures precision and recall across existing tools using a curated ground-truth of AST node mappings. This infrastructure supports rigorous comparisons and enables deeper investigations into the impact of AST representations and algorithm design choices.
Finally, we investigate the relationship between edit script length and diff quality by combining metric-based analysis with human feedback, revealing that minimizing edit length is not a reliable indicator of developer-preferred diffs
Spatial distribution of wildlife road mortality: How important is rigorous data collection?
No full textWe conducted 66 road mortality surveys along a 31.5 km stretch of a high-traffic 4-lane highway between Montreal and Sherbrooke in Québec, Canada. Surveys by vehicle between May and August 2019 recorded 212 animal carcasses from 48 species. Hotspots and coldspots for ground-dwelling vertebrates (mammals, amphibians, and reptiles) and birds were identified at four scales. The number of hotspots was higher at finer scales, while the combined length of hotspots was greater at coarser scales. We found significantly more animals and identified more species than the highway patrol, suggesting that rigorous road mortality surveys are beneficial for better understanding road impacts on biodiversity. We estimated that the highway patrol’s reporting probability for medium-sized mammals (more than 0.65 kg, less than 30 kg) was between 21 % and 54 % that of our systematic surveys. We recommend priority locations for mitigation to reduce road mortality and re-establish connectivity between wildlife populations separated by the highway
Mortalité animale sur les routes laurentiennes : Comment les collisions faune-véhicule sont-elles liées à l'utilisation des structures de passage par la faune ? Bulletin d'information 1
No full text- Identifier et évaluer les cas de mortalité faunique et l’utilisation des ponceaux le
long de la Route 117 (R117) et du Chemin du Lac-Supérieur (Ch. L.
-S.) dans les Laurentides
- Les inventaires de l'été 2025 ont révélé: 249 animaux tués sur la R117, et
1864 animaux tués sur Ch. L.-S.
- La mortalité des mammifères était plus élevée chez les écureuils roux, les lièvres
d’Amérique et les cerfs de Virginie.
- La mortalité des amphibiens était extrêmement élevée, avec 1571 individus,
dont plus de 1000 grenouilles vertes.
- Le suivi comprend 16 ponceaux observés par 74 caméras
A Poverty of the Soul: History, Impact and Assessment of the Prosperity Gospel in the United States
Full text linkThe Prosperity Gospel is not a gospel. It has not been subjected to the process of canonization, consequently, it will not be found in the Bible. It is a term pejoratively ascribed to evangelical Christian groups who believe that God wants them to be wealthy and that their wealth is a sign of God’s favour. Obtaining this wealth appears to be contingent on the strength of one’s faith. However, a yard stick of that faith can only be determined by how wealthy or poor a person is or will become. A wealthy person in theory has a greater faith than a poorer person. Acting on the claimed premise that the Bible is inerrant, Prosperity Gospel adherents use a fundamentalist approach in their hermeneutic. However, their area of concentration is mainly with Scripture that advocates wealth and riches. Although there are ample verses in the Bible to support this point of view, PGAs embrace a particular hermeneutic. This thesis will examine the Prosperity Gospel, how it evolved, its purpose, who are its adherents, who are its leaders, and any impact it may have on the social reality of the United States
Leveraging Machine Learning Classifiers for Backorder Prediction: A Comprehensive Framework for Enhancing Supply Chain Efficiency and Inventory Management Addressing Class Imbalance issue in Backorder Prediction
Full text linkThe current research explores the application of advanced machine learning and ensemble learning techniques to address the challenges of backorder prediction in supply chain management, specifically when the dataset is severely imbalanced. Considering the critical importance of accurate forecasting in supply chains, this study evaluates the performance of five resampling techniques (Random Under Sampling, ADASYN, SMOTE-ENN, Borderline-SMOTE, and SMOTE-SVM), combined with hyperparameter tuning (Randomized Search CV) and two cross-validation methods (5-fold and 10-fold). The research methodology involved training 98 combinations of two machine learning and five ensemble learning models, incorporating feature selection with SHAP and dimensionality reduction using PCA, alongside sophisticated data preprocessing techniques such as MICE for handling missing values. The primary evaluation metric is AUC-ROC, complemented by secondary metrics including balanced accuracy, F1 Score, and AUC-PR, ensuring a holistic assessment of model performance. Key findings demonstrate that ensemble learning models, particularly XGBoost, outperforms classical machine learning models in terms of robustness and being accurate in backorder prediction. Resampling techniques such as SMOTE-ENN and Random Under Sampling significantly enhance model performance, with SMOTE-ENN proving especially effective due to its noise reduction capabilities. Interestingly, dimensionality reduction using PCA was found to have little benefit, whereas feature selection using SHAP consistently improved efficiency and accuracy. The insights derived from this study provide a comprehensive framework for improving predictive performance in supply chain management applications, specifically backorder prediction. By addressing class imbalance, optimizing preprocessing techniques, and rigorously evaluating resampling methods, this research establishes best practices for tackling forecasting challenges in imbalanced, high-dimensional data environments
PROPRIOCEPTION IS THE CRITICAL TERM THAT KNOWS ITS SELF AS THE MID-THING
No full textThis thesis investigates an emerging “literary proprioceptive discourse” by connecting contemporary uses of proprioception as a critical term to Charles Olson’s 1965 poem-essay “Proprioception.” Olson’s “Proprioception” figures proprioception as a mid-thing between subject and object. Within the logic of his projective framework, the proprioceptive reader toggles between making and becoming, a modality wherein composition operates as a form of reception and vice versa. Those subject-object affordances that Olson illuminates back in 1965 anticipate contemporary usages and treatments of proprioception as a critical term in literature. In the yet to be defined proprioceptive discourse, there is little agreement in how to understand proprioception. Some scholars treat proprioception as an approach to reading while some understand it as an object of interest. But almost across the board, there is the sense that proprioception as a critical term affords a window in to understanding how text objects bear on a reading subject, and vice versa. Within these frameworks, integrating proprioception into the reading act works as a mode for understanding the slippery subject-object relationship of reading
Corporate Strategic Leadership and Hedge Fund Activism
No full textOver the past decades, activist hedge funds have come to the forefront of shareholder activism. Often viewed as proactive defenders of shareholder primacy, activist hedge funds acquire ownership stakes in corporations they perceive as mismanaged and advocate for changes to enhance shareholder value. This dissertation examines underexplored aspects of hedge fund activism through three essays, using a longitudinal dataset of companies listed in the Standard & Poor’s (S&P) 500 index between 2000 and 2020.
The first essay, grounded in corporate strategic leadership research, investigates whether social evaluations of CEOs—fame, celebrity, and infamy—affect the likelihood of their firms being targeted by activist hedge funds. Findings suggest that firms attract increased activist attention when their CEOs receive extensive media coverage, achieve celebrity status, or become infamous. I also find that this relationship is more substantial for firms led by female CEOs.
The second essay explores the impact of hedge fund activism on board gender diversity in both targeted and non-targeted firms. Results indicate that activism is associated with declines in board gender diversity in targeted firms, and that this negative effect is mitigated when a celebrity CEO leads the company. Contrary to expectations, no evidence of spillover effects on non-targeted firms was found, and potential explanations for this finding are discussed.
The third essay examines why some firms respond to hedge fund activism with hostile resistance by adopting defensive governance provisions, such as shareholder rights plans. Drawing on upper echelons theory, I argue that CEOs’ political ideologies and power shape corporate resistance strategies against hedge fund targeting. Empirical findings partially support this argument, revealing a strong positive relationship between CEO power and the likelihood of hostile resistance.
Together, these essays contribute to the literature on shareholder activism and corporate strategic leadership, offering new insights into how CEO characteristics and governance dynamics shape firms’ interactions with activist hedge funds