21793 research outputs found
Sort by
Let’s Share! Can Anthropomorphic Storybooks Promote Pro-Social Skills?
No full textThis study analyses whether different types of characters (human versus animal) in children’s picture books play a significant role in children’s sharing behaviours and reading comprehension skills, depending on the type of reading method used (standard versus dialogic). I examined children’s sharing behaviours using a sticker test before and after reading sessions. A total of 44 children participated in this study between the ages of 4 and 6 years old, who were randomly assigned to one of four experimental groups. Participants were read their assigned storybook a total of three times, once a day, three days in a row. After analyzing the collected data, no significant differences were found in sharing behaviours regardless of the reading group the participating children were in. However, differences in reading comprehension were present. There was a significant difference between the animal-standard group and the human-standard group. In sum, my study suggests that storybooks containing animals as protagonists do not necessarily decrease sharing behaviours and could improve reading comprehension in young children.
Keywords: Anthropomorphism, dialogic reading, reading comprehension, sharin
Participants’ Perceptions of a Receptive Raga Music Therapy Experience: An Interpretative Neurophenomenological and Arts-Based Analysis
No full textFor years, Indian musicians, scholars, and music therapists have explored the therapeutic use of raga. Research is emerging on its use in complementary medicine in India, yet little formal research exists on the use of raga in receptive music therapy with clients unfamiliar with this style of music. This mixed methods study used interpretative phenomenological analysis, neurophenomenological analysis (through electroencephalogram brainwave recordings via a Muse 2 headband and Mind Monitor software), and arts-based hermeneutic analysis to investigate the experiences of six participants unfamiliar with raga during a recorded receptive raga music therapy experience. Four group experiential themes emerged from the coding and analysis of interviews: (a) curiosity and interest with Indian classical music and culture, (b) imagery and pre-associations despite musical unfamiliarity, (c) increased immersion as musical movements progressed, and (d) the music allowing a change of consciousness. Both group-level and individual experiential themes were examined while acknowledging the rich nuances within the participants’ individual experiences. Brainwave results generally indicated an increase in alpha waves (associated with relaxed, meditative states) or delta waves (linked to deep meditative or sleep-like states), along with a decrease in beta waves (associated with alertness). A comprovisation approach using the qualitative findings and electroencephalogram data for two diverging participants offered a more embodied and comprehensive understanding of the phenomenon. The results were integrated in the discussion. This study contributes to an intercultural and multimodal understanding of raga in receptive music therapy and explores potential applications in guided imagery and music practices
Community, Identity, Belonging, and Jazz: An Exploration of Montreal’s Jazz Scene
No full textThis autoethnographic study examines the notions of identity, belonging, and community formation within the Black Anglophone community in Montreal. For my research, I understand Jazz as more than a genre that encompasses Hip-Hop, Blues, R&B, and other related styles, to develop an inclusive and holistic understanding of the musical and educational subcultures formed under the umbrella of Jazz. I employ music as a form of data collection, analysis, and discovery, by engaging in narrative discourse mixed with the visual and qualitative methodology of photography, field notes and music memoing and narrative vignettes. I used thematic analysis to explain themes that emerged throughout the research. The main findings of the study offer critical insights into how Jazz can transcend space, how self-exploration as a Black Anglophone provides a lens to understand belonging within the community, and how the improvisational nature of Jazz is connected to resilience, which has helped build and maintain the Black Anglophone community in Montreal
SpokenWeb Metadata Schema and Cataloguing Process
No full textSpokenWeb Metadata Schema has been developed by a group of researchers who are dedicated to the discovery and preservation of sonic artifacts that have captured literary events of the past to activate these artifacts in the present. The schema was developed as a part of the SSHRC-funded SpokenWeb partnership, and implemented as a schema in Swallow Metadata Management System and the SpokenWeb Search Engine. This is Version 4. The SpokenWeb schema describes both literary activities or events and the AV assets or digital files that document them. The instructions provide details about the information to be entered in each field
Early Layer Optimization
Get PDFIn deep learning, early layers play a fundamental role in building general and transferable representations. In this thesis, we demonstrate how improving early layer features can consistently enhance generalization across diverse training settings.
First, we propose a novel iterative training method called Simulated Annealing in Early Layers (SEAL), which applies intermittent gradient ascent followed by descent to the early layers during training. This enables the early layers to escape local minima and refine their representations over time. Doing so reduces overfitting leading to state-of-the-art in in-distribution and transfer generalization in iterative training regime.
Second, we observed poor transfer generalization in greedy learning which we attribute to the lack of generic information especially in the early layers of the network. To address this, we utilize CS-KD regularization to encourage information gain in the early layers. Our results show that this adjustment mitigates the transfer performance drop typically observed in greedy training, while maintaining in-distribution accuracy.
Finally, we extend our investigation to federated learning, where early layer divergence due to gradient accumulation across clients can lead to poor representation learning, even under IID data distributions. We demonstrate that greedy training, by avoiding end-to-end backpropagation, mitigates divergence in the early layers and improves overall performance, particularly in challenging scenarios with deeper models or many clients.
Overall, this thesis highlights the importance of early layer learning in building models that generalize well, and introduces practical strategies for improving it across iterative, greedy, and federated learning paradigms
Shifting the Focus: A Study of Student-Centered Learning in Modern Education
Get PDFThis thesis explores the world of education and the methodologies surrounding it by asking students and professors for their opinions on how they perceive academia and the best ways to teach and learn. It focuses on the role of alternative approaches, specifically Student-Centered Learning (SCL), to assess whether implementing these methods in universities would be beneficial. SCL proposes a shift in roles, with the professor transitioning from a "sage on the stage" to a "guide on the side". This research explores university teaching methods through online surveys and interviews with students and professors, aiming to understand their opinions on SCL. The goal is to understand if students obtain what they want and need through their academic journey. It also looks at professors' understanding of their roles within this context. The primary focus of this thesis is SCL, seeking to demystify students' understanding and their experience with this teaching approach to understand if it should be implemented more frequently in undergraduate university programs. By examining dropout rates, students' biggest challenges, key transferable skills needed for success, readiness for higher education, and more, this thesis provides a holistic understanding of what students like and dislike about the university and offers suggestions for the potential improvement of academia. This research reveals that half of students are ill-prepared for university. Furthermore, it emphasizes that flexibility is key to enhancing student success rates. Many participants in this study regard SCL as an effective approach to providing the adaptability necessary for improved academic outcomes
When Noise Helps: From Impossibility to Approximate Impossibility in Identity Effects for Deep Learning
Get PDFArtificial intelligence, and especially deep learning, have had a major spotlight shined on them in recent years. Our goal is to mathematically and computationally explore some of its potential limits - specifically, those related to identity effects and generalization. Identity effects refer to the ability to recognize specific patterns, such as checking whether two sounds are identical, something human brains are very adept at handling from a very young age. It has been observed that depending on how the input data is ``encoded”, that is what type of vectors differentiate the pair of objects to be classified a deep neural net might fail to generalize identity effects outside the training set. Namely, given that matching objects are valid pairs, and non-matching objects are not, some types of encodings will allow the network to generalize to inputs it has never seen before, and some will not. Previous research shows that the existence of a specific orthogonal transformation on the encodings can predict impossibility of generalization. We explore mathematically and then test numerically whether approximate orthogonality leads to predictable loosening of the impossibility, eventually finding that approximate orthogonality leads to approximate impossibility. In particular, adding random noise to traditional encodings such as the ``one-hot'' encoding can help neural networks generalize outside the training set
Advancing Cybersecurity in EV Charging Infrastructure: Vulnerabilities, Attacks, and a Real-time Monitoring and Detection Platform
Get PDFElectric Vehicles (EVs) have gained significant popularity as a sustainable alternative to traditional vehicles, leading to the rapid deployment of EV Charging Stations (CSs). While vulnerabilities have been discovered in CSs with remote management portals, those requiring physical access are considered secure and have received less attention. This thesis introduces the first attack framework targeting these CSs that require physical access to operate their local management portal. By testing six real-world CSs deployed across North America and Europe, the thesis demonstrates how attackers can exploit design flaws to escalate privileges, manipulate configurations, and launch attacks that disrupt the power grid, enable financial fraud, and compromise charging infrastructure availability. In response to these vulnerabilities and cyberattack threats, the research develops EV-Shield, a real-time monitoring platform that collects and correlates data from multiple EV ecosystem components including CSs and the EV Charging Station Management System (CSMS). The platform is integrated with HydraEV, a distributed anomaly detection module that enables attack detection through its local and central detectors. By incorporating Smart Meters (SMs) as a novel monitoring component, EV-Shield enhances its resilience, ensuring reliable detection even when other data sources are compromised. The platform's performance is evaluated through multiple attack scenarios, demonstrating its effectiveness in safeguarding the public EV charging infrastructure. This thesis addresses critical security challenges in the rapidly evolving EV ecosystem, focusing on CS vulnerabilities and developing a real-time monitoring platform to detect cyberattacks. The goal is to ensure the safe and reliable expansion of EV infrastructure amid growing cybersecurity threats
Harnessing Reconfigurable Intelligent Surfaces For Next-Gen Wireless Networks: Enhancing Efficiency, Reliability, and Security
Get PDFAs wireless networks evolve to meet the demands for high-speed, reliable, and secure communications, emerging technologies like Reconfigurable Intelligent Surface (RIS) are set to reshape wireless environments. This thesis investigates the transformative role of RIS in next-generation networks, focusing on performance enhancement and security. The research is divided into two
major contributions.
The first part examines the integration of an active STAR-RIS with a full-duplex Cooperative Rate Splitting Multiple Access (FD C-RSMA) system in a downlink Multiple Input Multiple Output (MISO) configuration. To tackle the non-convex optimization problem, an alternating optimization
framework based on successive convex approximation (SCA) is developed, achieving up to a 20.3% improvement in network sum rate. Additionally, to simplify real-time decision-making, a deep
reinforcement learning model using an actor-critic architecture is proposed, reducing computational time by 98% compared to the SCA method.
The second part explores the effects of a movable antenna (MA)-assisted jammer in a downlink MISO system. The analysis shows that MA-based jamming causes a 30% reduction in sum rate and a 25% increase in outage probability compared to fixed antenna setups. Furthermore, the study evaluates the effectiveness of RIS as a countermeasure against such adversarial attacks under different levels of jammer knowledge. Safeguarding RIS channel state information is found to be critical, as its compromise renders the system ineffective.
Overall, this research provides a comprehensive framework for utilizing RIS to enhance communication performance and strengthen security in wireless environments, laying the foundation for robust next-generation networks
An Ontology-Based Model for In-Network Computing Components Description and Discovery
Get PDFIn-Network Computing (INC) refers to the process that enables the distribution of computing tasks across the network instead of computing on servers outside the network. Advances in programmable hardware allow computations directly within network devices, such as switches and Smart Network Interface Cards (smart NICs), as data passes through them, and thereby reduces network congestion, minimizes reliance on distant cloud servers, and improves latency.
The growing demand for ultra-low delays, high bandwidth, and the ability to dynamically synchronize data streams in emerging applications, particularly Holographic-Type Communication application, is pushing the limits of current network infrastructures, with INC emerging as a promising solution as it aims to meet these stringent requirements. In addition, efficient provisioning of INC requires a comprehensive understanding of INC components, including their specifications, configuration information, and requirements. Nevertheless, an architecture with a description and discovery model offers a structured and standardized framework to represent INC components, defining their functionality and characteristics, and retrieving the most pertinent INC components according to the user requirements.
This thesis proposes a novel architecture for describing and discovering the most relevant INC components based on user preferences, specifically tailored for holographic-type application. The contribution of this work is threefold. First, we introduce the In-Network Computing Ontology (INCO), a domain-independent, ontology-based description model that provides a semantic representation of INC components, facilitating their discovery from a centralized repository. The description model covers both the functional and non-functional specifications of INC components and consists of two parts: a generic description for INC components and an extension detailing four specific INC components (i.e. encoder, decoder, transcoder, and renderer)- essential for our holographic application use case. Second, we present a semantic matchmaking algorithm that leverages the INCO model to automatically identify and select the most appropriate INC components based on user requests and preferences. Lastly, we validate the proposed approach through experimental simulations, demonstrating the algorithm’s effectiveness in terms of response time and consistency. Response time was measured based on two criteria: query complexity and the number of retrieved instances. The simulation results indicate that response time fluctuates with increasing query complexity, while it remains comparatively stable as the number of retrieved instances grows