IRis Bishop's University
Not a member yet
114 research outputs found
Sort by
Meta-Unsupervised Learning: Application to Non-Negative Matrix Factorization
Full text linkMeta-learning was initially developed for supervised learning to enable models to generalize across tasks by leveraging prior experience. However, its potential in the unsupervised learning domain remains underexplored. To demonstrate the feasibility of meta-learning in unsupervised settings,we apply it toNon-negative Matrix Factorization (NMF), a widely used technique for decomposing non-negative data matrices into interpretable, lower-dimensional representations. While NMF has found applications in topic modeling, image processing, bioinformatics, and recommendation systems, it faces persistent challenges such as rank selection, optimization stability, uniqueness, and computational efficiency. Traditional approaches primarily focus on improving initialization strategies to enhance convergence and generalization. However, these methods often fail to exploit structural similarities across tasks. This paper introduces a meta-learning paradigm for NMF that systematically learns optimal factorization parameters from small-scale tasks and transfers this knowledge to improve learning on larger tasks. By discovering fine structures in small tasks and leveraging them to guide factorization on more complex datasets, our approach directs the search process toward a more optimal and structured search space, reducing the risk of suboptimal solutions and improving model robustness. This meta-unsupervised learning framework enhances NMF’s ability to uncover meaningful patterns while maintaining adaptability across different domains. Additionally, we evaluate the model under noisy conditions and demonstrate its robustness by filtering noise over learning epochs, further enhancing its interpretability and stability. By integrating meta-learning principles, our method improves optimization stability and enhances interpretability and generalizability, bridging the gap between NMF-based models and advanced autonomous (unsupervised) learning strategies. The source code of this work is available at https://github.com/akhan232/meta-nmf.© Ameer Ahmed Khan, 202
Le service national de dépôt partagé Scholaris: Retours d’expérience de 3 universités québécoises
Full text linkLe libre accès s’impose comme un modèle incontournable dans le paysage de la communication savante. Les dépôts institutionnels jouent un rôle crucial pour assurer la diffusion et la pérennité des travaux de recherche. Cependant, l’implantation et la gestion d’un dépôt institutionnel peuvent représenter un défi de taille. C’est dans cette optique que Scholaris, un service de dépôt partagé canadien fonctionnant avec le logiciel libre DSpace, a été développé. Développé par l’Association des bibliothèques de recherche du Canada (ABRC), le Conseil des bibliothèques universitaires de l’Ontario (OCUL) et les bibliothèques de l’Université de Toronto, en partenariat avec Scholars Portal et des experts, Scholaris offre un hébergement centralisé, des mises à jour coordonnées et une localisation sécurisée des données au Canada. Cette communication présentera les retours d’expérience de trois universités québécoises – l’Université de Montréal (grande institution), l’Université de Sherbrooke (taille moyenne) et l’Université Bishop’s (plus petite) – en tant qu’utilisatrices précoces de Scholaris. Ces établissements illustrent comment une technologie partagée peut s’adapter à des contextes variés tout en optimisant la gestion des dépôts
Physiological Correlates of Graymatter
Full text linkThis thesis explores the physiological correlates of gray matter by correlating heart rate (HR) and heart rate variability (HRV) with cortical volume and thickness across various brain regions. Using six open datasets from open- Neuro comprising over 1000 healthy human subjects we first computed HR and HRV across all subjects using the photoplethysmography (PPG) signal. We then used the popular neuroimaging tool FreeSurfer to segment and quantify the brain’s gray matter based on T1-weighted images, extracting gray matter volume, thickness, and curvature from 31 distinct bilateral cortical regions. Finally, we computed the pearson correlation coefficient between our physiological metrics (HRV and HR) and gray matter metrics (volume, thickness, and curvature). When pooling all regions and all subjects we find a significant inverse correlation between gray matter volume and HR (more gray matter predicts lower heart rate), a positive correlation between HRV and gray matter volume, inverse correlation between HR and thickness, and a very robust positive correlation between HRV and thickness (subjects with higher HRV also had thicker gray matter). There was no relationship between gray matter curvature andHRV or HR. The correlations between HRV and cortical thickness were strongest in the medial orbitofrontal and pars opercularis brain regions, suggesting a potential link between autonomic nervous system regulation and structural aspects of brain regions involved in emotional, cognitive, and social functions. This is the first study to comprehensively combine physiological and MRI-based gray matter data from over 1000 healthy subjects, advancing our understanding of the brain-body connection and its role in health, emotion, and cognition.© Mohammad SohrabiGharehtappeh, 202
Enhancing Key Press Motor Task Classification in Brain Computer Interfaces Using an Integrated Hybrid Deep Learning Model for EEG Signal Processing
Full text linkEEG signal classification plays a pivotal role in a diverse array of biomedical applications, particularly in the development and optimization of brain-computer interfaces (BCIs). Accurate classification of EEG signals is essential for translating neurological activity into meaningful commands for users, thereby enhancing the efficacy of BCIs in real-world scenarios. In this research, we propose a deep learning model based on Long Short-Term Memory (LSTM) networks, termed LSTM_SST, which adeptly integrates spatial, spectral, and temporal features derived from EEG signals to improve the classification of key press motor tasks. The LSTM_SST model is designed to effectively capture the temporal dynamics of EEG activity while simultaneously harnessing spatial and spectral information from EEG topographic maps and power spectral density estimates. By bringing together these different features, the model aims to provide a comprehensive representation of the underlying EEG signals, crucial for accurate classification. Our study rigorously compares the performance of the LSTM_SST model against several established models, including EEGNet, ShallowConvNet, graph based DCRNN (Dynamic Control Recurrent Neural Network), and LSTM architectures. The evaluation metrics employed in this research include classification accuracy, precision, recall, and F1-score, ensuring a comprehensive assessment of model performance. The results indicate that the LSTM_SST model not only achieves competitive classification metrics but also shows particular promise in effectively differentiating between the various key press tasks. The integration of spatial, spectral, and temporal information within the LSTM_SST model allows it to capture a more holistic viewof EEG signals, thereby enhancing its classification capabilities.© Abass Zakari, 202
On TorXakis Correctness as an ioco Implementation: An Empirical Model-Based Evaluation
Full text linkThe ultimate goal of this thesis is to work toward the use of model-based testing techniques to evaluate whether TorXakis is a correct implementation of the inputoutput conformance (ioco) testing theory. Rather than pursuing a formal proof of correctness, we adopt an empirical, model-based testing approach to evaluate whether TorXakis adheres to the expected semantics defined by ioco. A wide range of custom test models were designed and executed, each targeting specific system behaviors such as concurrency, synchronization, fault tolerance, and deadlock handling. These models simulate real-world challenges to assess whether TorXakis produces outputs and traces that align with the theoretical behavior prescribed by ioco. Through structural, behavioral, and trace-based analysis, we collect evidence that supports or challenges TorXakis’s conformance to ioco principles. The tests were conducted under controlled conditions, systematically increasing complexity to expose potential deviations or inconsistencies. While the study does not offer a formal verification of TorXakis, it provides a practical and meaningful evaluation that lays the groundwork for future formal investigations.@ Reza Ghasemi, 202
Supporting Student Teachers in Practicum: Associate Teachers’ Experiences of Wellbeing and Relational Care
Full text linkThis thesis explores how associate teachers perceive, support, and experience student teacher wellbeing during the teaching practicum. While the practicum represents a crucial phase in teacher education, it often places emotional demands on both student teachers and the associate teachers who mentor them. The study responds to two interconnected gaps: an empirical gap in understanding how associate teachers engage with wellbeing in practice, and a conceptual gap in how teacher education frameworks define and position wellbeing within mentoring relationships.
Using an explanatory sequential mixed methods design, the research integrates survey data from 23 associate teachers with in-depth interviews from four participants working in English-language elementary schools in Quebec. The findings reveal that associate teachers tended to view wellbeing as relational and dynamic. They often relied on informal, intuitive practices to support student teachers, navigating unclear expectations about their dual roles as mentors and evaluators within systems that provide limited institutional guidance.
The study highlights that meaningful support for student teacher wellbeing depends on recognizing and resourcing the mentorship role. Key implications include reframing mentorship as relational pedagogy, strengthening communication within the practicum triad, and embedding wellbeing into the structure of teacher education programs. Overall, the research calls for a more integrated and humane approach to mentorship that treats wellbeing as a shared and systemic responsibility rather than an individual burden.© Theresa Gagnon, 202
“What’s the point of it all if we are just going to die anyway?”: A qualitative exploration on arts-based existential interventions as a tool for navigating children's concerns about death
Full text linkBackground: Navigating the topic of death is no easy task for most, particularly when children are involved. While some adults fear that discussions on death with children may cause distress, research suggests that they are naturally curious and eager to learn more about it (Paul, 2019). However, parents and educators often feel uncomfortable and unqualified to lead these dialogues, leaving children without supportive spaces to explore it openly. Study Purpose: This thesis aimed to address this concern by using a combined arts-based and philosophical intervention—termed arts-based existential interventions—to support children and their concerns regarding death. Methods: A 10-week intervention was conducted with two sixth-grade classrooms (n=21, Mage 11.5) using a descriptive qualitative design. Two of those ten workshops—focused on personal mortality and bereavement, respectively—formed the core of this thesis. Each workshop included an art activity followed by a semi-structured philosophical inquiry. Data was collected through audio recordings, observation grids, artwork analysis, and semi-structured interviews. Results: Our findings indicate that the intervention provided a worthwhile space for children to express their thoughts and emotions about death. Analyses revealed two distinct coding trees that formed the basis of two separate research articles, presented in chapter 2 of this thesis: (1) a grounded theory analysis linking participants’ reflections to Martin Heidegger’s concepts of inauthenticity, authenticity, Being-towards-death, existential anxiety, and anticipatory resoluteness; (2), a thematic analysis examining the general appreciation of arts-based existential therapy on the mental health of youth, particularly as it pertains to their navigation of death. Following this chapter, in a general discussion, we elaborate on arts-based existential interventions for fostering children’s mental health. Implications: Findings suggest that arts-based existential interventions can help cultivate children's emotional resilience toward mortality. Future research should expand on this intervention by exploring additional existential themes such as suffering, empathy, meaning in life, and freedom.@ Zachary Fry, 202
Heroes Arena: A Low-Latency Architecture for Fully On-Chain Card Battle Games
Full text linkFully on-chain games often suffer from significant interaction delays due to blockchain transaction latencies, hindering the user experience, especially in real-time gaming scenarios. This thesis presents Heroes Arena, a fully decentralized real-time card battle game deployed on the Ethereum Sepolia testnet, designed to minimize user-perceived latency. By employing a parallel architecture that loads all necessary blockchain data upon user login and subsequently executes all gameplay logic locally, this system offers immediate in-game feedback without additional blockchain confirmations. Game events and rewards, such as NFT distribution and in-game resources, are asynchronously queued and committed to the blockchain in order, ensuring consistency without interrupting player interaction. The complete game implementation—including real-time battles, NFT card management, resource purchasing, and timed reward mechanisms—was independently developed using Unity and Solidity smart contracts. Comparative analysis demonstrates that the proposed architecture achieves significantly improved responsiveness and simplicity compared to existing zero-knowledge proof-based methods and ephemeral rollup solutions.© Haoqing Hu, 202
Everyday Diplomats: The Global Solidarity Movement for Timor-Leste, 1975-99
No full textSolidarity activists with Timor-Leste (East Timor) during the period of Indonesian military occupation (1975-99) were involved in a form of diplomacy. It was messy and often loud; an unconventional approach compared to state diplomacy. Yet the movement emerged as an effective tool for the Timor-Leste cause, and as a constructive form of “other diplomacy” that often proved influential in shaping both government and Timorese actions. It legacy lingers even after Timor-Leste’s independence
Enhancing Data Center Reliability Through Deep Learning-Based Disk Failure Prediction
Full text linkWith the growing reliance on technology, data storage and failure prediction have
become critical and challenging issues. Among various data storage methods, disks
are the most commonly used components in storage systems, with the majority of
information currently stored on them. As a result, disk failures can lead to irreparable
damage. Although such failures are relatively rare, large-scale storage systems
containing thousands of disks are still prone to severe failures, often resulting in
permanent data loss. For this reason, maintaining the reliability of storage resources
has always been a serious challenge. Various methods have been developed to detect
and predict disk failures in data centers. However, identifying failures quickly
and accurately remains a significant challenge. Due to the poor performance of old
methods, researchers are motivated to use different techniques to detect failures
earlier and more accurately to cover the weaknesses of the old methods. Deep
learning is one of the most advanced techniques used for predicting disk failures.
Thus, we train a Bidirectional Long Short-Term Memory (Bi-LSTM) deep neural
network to diagnose and predict disk failures effectively. In deep learning-based
failure detection methods, selecting effective features plays a crucial role in enhancing
the model’s accuracy and performance. However, using too many features
can increase computational load, add unnecessary complexity, and reduce overall
efficiency. To address this, we apply feature selection techniques as part of our
methodology. Specifically, we use Pearson correlation and Chi-square tests to identify
the most relevant features. The datasets used in this study are from Backblaze
and Baidu. The results demonstrate that our model accurately detects failures
with 98.36% accuracy and 97.8% precision, and successfully predicts failures up
to 40 days in advance. Additionally, we develop a decision tree-based model to
evaluate disk health status. This model predicts the remaining useful life of hard
disks and classifies them into Healthy, Warning, or Critical states. By classifying the
disk health features and then applying the trained Bi-LSTM, we achieve significant
improvements, reaching an accuracy of 99.27% and a precision of 98.65%.@ Yassaman Mardan, 202