eScholar OntarioTech University
Not a member yet
1901 research outputs found
Sort by
A retrieval augmented generation fine-tuned LLM model for refactored code recommendations to mitigate Java lock contention performance faults
Full text linkLock contention performance faults can lead to degradation in the performance of software applications. Unlike software bugs, performance faults do not lead to failures and application crashes but surface as a degradation in the response and execution of an application and can surface fairly late in the deployment life of an application. Tools exist for the identification and detection of lock performance faults but there is a lack of an effective code refactoring recommendation system for developers to mitigate the performance degradation caused by lock contention. Recent advances in Large Language Models (LLMs) have demonstrated positive results in code refactoring for fixing software bugs and mitigating runtime faults. However, traditional LLM-based approaches often suffer from wrong output and incorrect recommendations, where the generated code may not accurately reflect the context of the project or the existing codebase. This thesis presents a novel approach that combines Retrieval Augmented Generation (RAG) with a fine-tuned LLM model for refactored code recommendation aimed at reducing lock-contention performance faults in Java applications. The RAG-based fine-tuned model combines the strengths of contextual understanding from LLMs with the precision of retrieval-based systems, thereby ensuring that the generated recommendations are relevant, accurate, and domain-specific. Semantic and syntactic metrics of the recommendations generated by the combined RAG and LLM model show an accuracy of approximately 90% compared to an accuracy of approximately 25% when a baseline LLM model is used
Structure-based learning via graph neural networks for multi-group multicast beamforming
Full text linkIn this thesis, we consider a downlink multi-antenna multi-group multicasting problem, where users in the same group request the same content. Even though existing optimization-based algorithms can obtain a sub-optimal multicast beamforming solution, they are designed under perfect channel state information (CSI) and need to be performed each time channel state information is updated. Due to the complexity involved in solving the problem, they lead to high computational run-time for largescale antenna-array systems. To overcome this challenge, we propose a robust efficient learning-based approach for a scalable beamformer design under imperfect CSI. We mainly focus on the design of the base station (BS) multicast beamformer to maximize the weighted signal-to-interference-plus-noise ratio (SINR) among all users, subject to a transmit power budget constraint, also known as the max-min fair (MMF) problem. We propose to solve the MMF problem by developing a graph neural network (GNN) model and explore the available optimal multicast beamforming structure to speed up the training process and improve performance. A scalable GNN-based beamformer architecture critically depends on the design of the hidden layers. By exploring the interactions between user groups, we construct the hidden layers of our proposed GNN to effectively capture intra-group and inter-group interaction patterns, enabling the network to learn the beamforming solution effectively. Our GNN via design can also generalize different numbers of users and groups without re-training. Simulation results show that the proposed GNN model achieves a near-optimal performance. We also show that our GNN-based model is significantly faster compared to conventional optimization-based methods and outperforms other existing learning-based models. It can be well generalized to different numbers of users and groups with excellent accuracy
Library Research Skills for Knowledge Synthesis
No full textThis course has three modules that are designed to be completed independently at your own pace. Each module will take approximately 1 hour to complete.
Module 1 provides an overview of the different types of knowledge syntheses and the steps involved in planning a review, such as how to develop your research question, identify databases and resources, and create and register a protocol.
Module 2 covers how to identify sources for your review. You will learn how to use keywords and subject headings, how to translate your search strategy to multiple databases, and advanced techniques for searching in databases and for grey literature.
At the end of each module is a quiz to assess mastery of the course learning outcomes. This course contains interactive learning activities for you to test your understanding before completing the quiz at the end of each module.
Module 3 discusses best practices for documenting your results and your searches to ensure transparency and reproducibility.This course is for any student who wishes to increase their proficiency in conducting research for systematic reviews, scoping reviews and other types of knowledge synthesis. Learn about different types of knowledge synthesis to identify the right type for your question, including systematic and scoping reviews. Find out how to navigate the process of completing a knowledge synthesis project, including planning, following a protocol, developing your search and retrieving and saving search results
Co-design and evaluation of a mHealth intervention to enhance dietary self-management and adherence in patients with heart failure
Full text linkMedical and dietary therapies (i.e., sodium and fluid restriction) are key to improving heart failure (HF) outcomes, a leading cause of morbidity and mortality in Canada. Despite support from healthcare providers (HCPs), ~50% of patients are nonadherent to these therapies. Mobile applications (apps) may support patient adherence to treatment; however, little is known about patient and HCP perspectives regarding the adoption of apps for HF management, which is critical to successful development and implementation. Moreover, a majority of existing HF apps have focused solely on symptom monitoring, and only a few have addressed dietary education and adherence.
This research conducted three studies. Two descriptive qualitative studies using focus groups and interviews assessed patient (Study 1) and HCP (cardiologists and nurses; Study 2) needs, motivations, and challenges related to using apps for HF management. Study 3 reported on the co-design, development, and user-testing (via quasi-experimental design) of a mobile app to support dietary education and adherence among patients with HF. App engagement, satisfaction, and usability were measured using a patient-reported questionnaire. Patients included in the research were outpatients with HF, ≥18 years, on stable medical therapy, and smartphone users.
Patients (n= 19) and HCPs (n=21) felt apps present opportunities for education (e.g., diet, medication, symptoms), patient engagement and communication, and enhanced efficiency of care (e.g., remote care). Factors such as time and workload impact on HCPs, age, and accessibility of technology were seen as challenges to app use. App features like reminders/alarms, quizzes, and gamification were favoured. These findings, integrated with behavioural and educational theories informed the co-design and development of a nutrition-focused mobile app by an interdisciplinary team of dietitians, researchers, and industry partner. The app included six educational modules on sodium and fluid restriction, ten behaviour change techniques (e.g., feedback on behaviour, goal setting) and gamified components (i.e., avatar, point-system). The app demonstrated a high level of usability, engagement, and satisfaction among patients with HF. Together, these findings support the importance of patient and HCP perspectives in the development and adoption of mobile apps and the potential of such tools to support dietary education and adherence for HF management
The impact of chiropractic manipulation on pain, functional outcomes and gait symmetry in dogs with mobility impairments using clinical measurement and gait analysis instruments
Full text linkMusculoskeletal (MSK) disorders are a leading cause of pain and mobility impairments in dogs, yet conservative treatment options remain limited. Animal chiropractic care is increasingly utilized to alleviate pain, improve mobility, and enhance athletic performance; however, empirical evidence supporting its efficacy is limited. This thesis employed a two-phase design to examine the effects of chiropractic manipulation on pain, functional abilities, and gait symmetry in client-owned dogs with mobility impairments. Manuscript 1 evaluated pain and functional status through owner-reported and practitioner-assessed questionnaires, while Manuscript 2 objectively analyzed gait symmetry using a pressure-sensitive walkway, evaluating primary temporospatial parameters including stride length, stance time, and total pressure index. Significant improvements were observed across pain, functionality, and gait symmetry, particularly at the trot, suggesting enhanced neuromechanical function and more balanced weight distribution. These findings provide preliminary evidence supporting chiropractic manipulation as a viable therapeutic option for canine MSK dysfunction, highlighting the need for further research
Quasimetric decision transformer: enhancing goal-conditioned reinforcement learning with structured distance guidance
Full text linkRecent works have shown that tackling offline Reinforcement Learning (RL) with a conditional policy produces promising results. Decision Transformer (DT) have shown promising results in offline RL by leveraging sequence modeling. However, standard DTs rely on Returns-to-Go (RTG) tokens, which are heuristically defined and often suboptimal for goal-conditioned tasks. In this work, we introduce Quasimetric Decision Transformer (QuaD), a novel approach that replaces RTG with learned quasimetric distances, providing a more structured and theoretically grounded guidance signal for long-horizon decision-making. We explore two quasimetric formulations: Interval Quasimetric Embedding (IQE) and Metric Residual Network (MRN), and integrate them into DTs. Extensive evaluations on the AntMaze benchmark demonstrate that QuaD outperforms standard DTs, achieving state-of-the-art success rates and improved generalization to unseen goals. Our results suggest that quasimetric guidance is a viable alternative to RTG, opening new directions for learning structured distance representations in offline RL
Comparative energy, economic, and environmental assessment of conventional and alternative fuel vehicles across Canadian provinces
Full text linkThis thesis evaluates passenger car powertrains and fuels across Canadian provinces in terms of environmental impact, economic cost, and energy efficiency. A life cycle assessment using TRACI methods assessed six environmental categories. Results show that FCEVs perform best in Manitoba (0.065 kg CO₂ eq/km), BEVs in British Columbia (0.07 kg CO₂ eq/km), LPG vehicles in Alberta (AP 0.013 mol H⁺ eq/km; RE 5.38 × 10⁻⁵ kg PM₂.₅ eq/km), and HEVs in Saskatchewan (EP 7.07 × 10⁻⁶ kg N eq/km; POP 0.0002 kg NOₓ eq/km). Economically, HEVs were most cost-effective in most provinces, with lifetime costs ranging from C105,496 (Alberta). BEVs also emerged as viable alternatives in Manitoba (C83,144). In terms of energy, BEVs achieved the highest well-to-wheel efficiency (30.37%), followed by HEVs (14.5%). Overall, BEVs are most suitable for Ontario, British Columbia, and the Atlantic Provinces; HEVs for Alberta and Saskatchewan; and FCEVs for Quebec and Manitoba
Misrepresentation of Islam and Muslims in Canadian media: a comparative content analysis of major news outlets
Full text linkThis study focuses on how mass media frames Islam and Muslims in Canada. This study addresses the rise in hate crimes and islamophobia which is often fueled by mass media’s misrepresentation of Islam and Muslims. Using quantitative and qualitative content analysis headlines were taken from The Toronto Star, The Globe and Mail, and National Post. The findings revealed that Muslims and Islam continue to be negatively portrayed negatively compared to other religious groups such as, Christians and Jews. Media headlines often generalize and perpetuate harmful stereotypes about Islam and Muslims, exacerbating their marginalization and “othering” in Western society. Positive media news analysis was conducted as well which revealed that while positive news addresses the Muslim problem, it fails to address the underlying issues that cause discrimination and hate against Islam and Muslims. The study concludes that unbiased and balanced reporting is needed to counter stereotypes and increase social cohesion
Examining the differentiating influences of socioeconomic status and adversity on executive functioning, social cognition and the rate of neural maturity in children and adolescents
Full text linkWe sought to disentangle the influence of socioeconomic status and adversity across six resting-state functional networks and the association between network maturity and cognition. We compared a sample (N = 216) of children and adolescents to adults by generating a neural maturity index, using independent component analysis and dual regression, to quantify the similarity of each participant’s brain to a sex-matched adult template. Sensory (sensorimotor) and association (default mode and executive control) networks maturated at a similar rate and sooner than other networks. Additionally, only the default mode and hippocampal networks were influenced by environmental factors. Maturity of the default mode was associated with less adversity and better social cognitive ability, whereas maturity of the hippocampal network was associated with younger participants with higher IQs. These effects were stronger in females compared to males. Our results highlight the importance of examining SES and adversity as distinct dimensions of childhood environments
Data-driven machine learning for simulating and predicting urban intersection traffic
Full text linkUrban traffic systems are hard to manage because quality data are scarce, realistic scenarios are difficult to build, and short-term volume forecasts are uncertain. This thesis investigates how machine learning improves urban traffic simulation and management, tackling poor data quality, realistic scenario creation, and reliable volume forecasting to bridge the gap between simulation and reality. It reviews existing simulators and models, surveys the use of Machine Learning in the context of urban traffic simulation, implements a DQN-based signal control system in Newcastle, Ontario that cuts average wait times by approximately 22%, introduces an open-source tool called CrossFlow to convert real world Turning Movement Count data into realistic SUMO scenarios, and analyzes several deep-learning architectures for traffic-volume forecasting on Toronto vehicular data. Validation across 106 Toronto intersections with varied data availability shows generalizable gains, indicating that adaptive, data-driven methods can improve urban traffic simulation and management