39339 research outputs found
Sort by
Device-free Human Activity Recognition for the Kitchen
No full textDevice-free human activity recognition (HAR) is a promising approach for smart environments, but existing methods often experience performance drift with context-specific tasks, particularly in complex settings like kitchens. While prior work explored coarse activities (e.g., walking, sitting), fine-grained, context-specific tasks, such as kitchen activities involving object interactions, remain widely underexplored and are challenging due to signal noise, environmental variability, and limited datasets. In this thesis, we begin by evaluating three HAR approaches for kitchen tasks: deep learning (43% accuracy, limited by small datasets), transfer learning (39%, hindered by domain mismatch), and classical machine learning (68%, which proved best suited for the context-specific data). Given the superior performance of classical learning, its reliance on effective feature selection becomes a critical challenge. Current methods often rely on heuristics to select features, which can limit performance and require domain knowledge. To address this, we propose an optimized method that combines multivariate analysis of variance (MANOVA) with genetic algorithms for discriminative feature selection. This approach improves accuracy to 80% for coarse-grained tasks such as storing food and washing pots. However, fine-grained motions (chopping and slicing potatoes) remain challenging (with average recognition accuracy of 37%), highlighting the limitations of WiFi sensing for subtle activities. The key contributions of this thesis include: (1) an evaluation of HAR methods for kitchen tasks, (2) a novel approach to feature selection based on MANOVA-genetic algorithm to generalize the feature selection process, and (3) a public WiFi-based dataset to advance research in context-aware HAR. These findings highlight the potential of classical learning in context-specific tasks while revealing open challenges in fine-grained activity recognition
Constructing New Families of Covering Arrays Using Finite Geometry and Finite Fields
No full textCovering arrays are well-studied objects in combinatorial design theory. A strength- \emph{covering array}, denoted by CA, is an array over an alphabet with symbols such that in any -set of columns, each -tuple over the alphabet occurs in at least one row. Finite fields and finite geometry have been widely employed in the construction of covering arrays. Let be a prime power. One of the significant geometric objects arising from \PG(2,q) that has been used to construct strength- covering arrays is a pair of \emph{orthogoval} projective planes. Two projective (affine) planes with the same point sets are orthogoval if the intersection of any two lines, one from each plane, has size at most two. The existence of such pairs of orthogoval projective planes has been independently proven and published multiple times.
In this thesis, we extend the concept of a pair of orthogoval projective planes to a key property involving M{\"o}bius planes in \PG(3,q). We say that (truncated) M{\"o}bius planes are \emph{anti-cocircular} if the common intersection of any choice of circles, one from each of the planes, has size at most three.
We prove the existence of three anti-cocircular truncated M{\"o}bius planes for any odd prime power . This new geometric object has significant properties that enable the construction of strength- covering arrays. The covering arrays obtained through this method significantly improve the upper bounds on the size of the best-known arrays. Moreover, these covering arrays have a rich structure, which can be beneficial to their use as ingredients in recursive constructions.
Our results suggest the existence of analogous properties in higher-dimensional projective geometries, with potential applications in the construction of higher strength covering arrays.
Another contribution of this thesis involves strength- covering arrays. Strength- covering arrays obtained from a pair of orthogoval projective planes have a significant property that makes them suitable for recursive constructions. Each such array is obtained by the vertical concatenation of two strength- orthogonal arrays that together form a strength- covering array.
Taking advantage of this structure, we construct new families of strength- covering arrays by first horizontally concatenating multiple copies of this strength- array. The coverage is then completed by adding key ingredient arrays, which together are used as part of a general recursive construction. In certain cases, the ingredient arrays are carefully selected to introduce systematic redundancy among rows, allowing redundant rows to be removed to optimize the size of the covering arrays.
These new families reduce the size of some of the best-known covering arrays. Such improvements were enabled by exploiting the diverse properties inherited from finite fields and finite geometry
Development of Photocatalytic Titanium Dioxide Shell on Titanium Powder for Cold Gas Dynamic Spray
No full textThe objective of the thesis was to produce a titanium dioxide shell on spherical pure titanium powder that had heterojunctions of anatase and rutile nanostructures with photocatalytic properties for cold gas dynamic spray applications. This would be the first step in the global project of sprayable titanium dioxide coatings for self-sanitizing surfaces. A heterogeneous structure would provide the stability and closer band gap to the visible light spectrum from rutile and the high photocatalytic reactivity from anatase. The presence of heterojunctions could increase the photo-reactivity by promoting charge separations and promote reactions under visible light with a smaller band gap. Cold gas dynamic spray would provide the mechanism to consolidate this material without unwanted phase transformations from anatase to rutile at higher temperatures. As the temperatures are below the metals melting point, a combination of ceramic titanium dioxide and ductile titanium would assist in successful deposition via cold gas dynamic spray.
Preliminary experiments were completed on readily available titanium plates to observe potential growths and compare the results to reference literature. The final powder production procedure included a hydrogen peroxide, melamine, and nitric acid chemical bath at 80 ℃ for twelve hours followed by calcination for one hour at 450 ℃. This experimental procedure yielded the growth of spiked nanorods with nanoflower aggregates seen through scanning electron imaging. X-ray diffraction verified the presence of anatase and rutile structures, and Raman spectroscopy verified the titanium dioxide shell was predominantly anatase near the titanium core, with anatase and rutile heterojunctions at the outer part of the titanium dioxide shell. Photocatalytic testing was completed by subjecting the titanium dioxide in deionized water to methylene blue, an organic dye commonly used as an indicator for evaluating photocatalytic performance, and an ultraviolet light source to produce reactive oxidizing species that accelerate the decomposition of methylene blue. The experimental powders photocatalytic reactivity was compared to that of two commercial powders, Altair and Neoxid. Methylene blue adsorption and subsequent decomposition by the powders was characterized through microplate light absorbancy readings. The experimentally produced powder performed comparable to the commercial powders, verifying its photocatalytic properties under ultraviolet light. Dynamic vapour sorption was completed on all three powders to characterize the surface areas and adsorption properties of the experimental and commercial powders to provide further understanding of photocatalytic results. The commercial powders had different particle size, specific surface areas, and dynamics of adsorption when compared to the experimental powder, which would impact photocatalytic behaviour.
Although the experimental powder did not have superior photocatalytic reactivity to those of the commercial powders, the motivation is the presence of a ductile titanium core would plastically deform below its melting temperature to promote deposition efficiency in cold gas dynamic spray applications and produce a successful photocatalytic titanium dioxide coating. Solely ceramic titanium dioxide powders, like Altair and Neoxid, experience brittle impingement and poor deposition from lack of plastic deformation and metallurgical bonding. Overall, successful growth of heterogeneous titanium dioxide shells was grown on titanium powder that possessed photocatalytic properties comparable to commercially available powders to be applied in cold gas dynamic spray applications. The work completed created a strong foundation for project development and expansion
Exploring Properties of MoS2 Transferred onto Periodically Patterned Substrates
No full textThis project aimed to characterize the strain in few-layer MoS2 via Raman spectroscopy. Periodic patterns were etched into Si/SiO2 wafers using photolithography, after which MoS2 flakes were exfoliated and transferred on top of the patterns, leading to a mechanical deformation of the layers. We investigated the topography of the flakes via atomic force microscopy, and discovered that the sudden drop-off of the flakes had led to an increased presence of air bubbles. By investigating our samples with Raman spectroscopy, we first confirmed the presence of the two characteristic peaks E1_2g and A1g. Comparison with other studies indicates that our MoS2 flakes are thinner than our AFM measurements suggested, which we attribute to presence of air bubbles, which can both artificially increase the step heights as well as disturb the vibrational modes. Modulations in the Raman response around the location of the bubbles shows a softening of the difference of the peak position, although the change is not significant enough. Similarly, no significant change in the Raman response was observed at the edges of the patterns, due to the shallow etch depth, which does not induce a strain important enough to be resolved by our Raman system. These results highlight the challenges associated with strain characterization of crystalline systems
Discourse on Emptiness: A Cognitive Model of the Two Truths Distinction in the Madhyamaka Philosophy of Nāgārjuna and Candrakīrti
No full textI treat the two truths distinction of the Madhyamaka school of Indian Mahāyāna Buddhism as a hermeneutic device for interpreting the philosophy of Nāgārjuna and Candrakīrti: I use the distinction to decipher what in their texts is conventional truth (saṃvṛti-satya) and what is ultimate truth (paramārtha-satya). Accordingly, the thesis I argue is that the two truths distinction correlates to a cognitive divide: I apply a cognitive reading to each of the two truths wherein each corresponds to a different mode of cognitive experience, and I formally differentiate the two through the criterion of the presence and absence of language and conceptuality therein. Candrakīrti defines a “conventional truth” and an “ultimate truth” each in terms of being an “object”. How each object is “seen” I correlate with distinct cognitions. The conventional truth (defined as comprising the objects of the valid cognition of ordinary experience) correlates with a conceptual, dualistic form of cognition; the ultimate truth (defined as comprising the “object(s)” of the valid cognition of extra-ordinary experience), correlates with a non-conceptual, non-dualistic form of cognition. Regarding its application, the formal differentiation corresponds to a twofold soteriologization. There is an interdependence between the cognitive reading of the two truths and their respective soteriological functions insofar as the cognitive distinction should be predicated by a twofold soteriologization. That is, beyond their formal differences, a difference in soteriological application distinguishes two modes of cognition. Each “truth” or form of cognition functions in its respective way as a soteriological device for attaining liberation and awakening. This cognitive-qua-soteriological distinction is significant for only through the realization of the ultimate is liberation attained. The conventional truth is insufficient to attain the soteriological goal; nonetheless, it is the means to achieving the ultimate and, therefore, that goal. This instantiates a specific relation between the two truths as one of means and goal. Any comprehensive model of the two truths must explain this progression in not only philosophical but soteriological terms
Automated Head Impact Detection in Youth Ice Hockey: A Two-Stage Deep Learning Approach
No full textDetecting head impact events in youth ice hockey is a critical problem at the intersection of computer vision, sports safety, and public health. Despite increasing awareness of the long-term neurological risks of repetitive head trauma, most youth leagues lack systematic monitoring. This thesis addresses that gap by developing and evaluating an automated detection system tailored to the resource constraints of community
hockey. The contributions are threefold. First, we introduce the first publicly available annotated dataset for this task, featuring a hierarchical labeling scheme for both general and head impact events. Second, we design a hierarchical two-stage pipeline for rare-event discovery: Stage 1 detects general impact events to prune routine play, and Stage 2 applies specialized classification to determine head involvement. Third,
we provide a comprehensive empirical comparison between a player-centric approach (Model A) and a full-frame multi-modal fusion approach (Model B). Experiments show that the optimal deployment strategy uses a player-centric architecture for Stage 1 and a full-frame architecture for Stage 2. The best Stage 1 model (Model A: TSM + Motion) achieves 75% recall and 25% precision (F1-score 0.375) for general impact detection, successfully reducing the portion of video requiring human review from 100% to just 6.3%. For the subsequent head impact classification task, the best Stage 2 model (Model B: RGB + Flow + Pose) achieves 80% recall and 67.6% precision (F1-score 0.733). This results in an end-to-end head impact event detection recall of 60%. Technical analyses reveal several key findings: motion (optical flow) is the dominant signal for impact detection; Temporal Shift Modules (TSM) consistently outperform Inflated 3D Convnets (I3D) in both accuracy and computational efficiency; and pose estimation features provide limited value for
this task due to low resolution and equipment occlusion. This work demonstrates the feasibility of automated safety monitoring in youth hockey and provides a foundational dataset and methodology for future research
Exploring the Safety and Efficacy of Regulatory T Cell Derived Extracellular Vesicle Therapy for Acute Respiratory Distress Syndrome
No full textAcute Respiratory Distress Syndrome (ARDS) remains a major cause of morbidity and mortality. Despite significant advances in critical care, therapeutic options are largely limited to supportive care strategies. Regulatory T cells (Tregs) and their extracellular vesicles (Treg-EVs) have emerged as promising immunomodulatory treatments, yet their application in ARDS remains critically limited. This thesis addresses this issue through three complementary aims: (1) identifying existing knowledge gaps using a systematic review of preclinical and clinical evidence on Treg-based therapies for acute lung injury (ALI, the preclinical counterpart of ARDS); (2) establishing and characterizing a clinically relevant, translationally validated viral ALI model; and (3) evaluating the safety and preliminary therapeutic efficacy of Treg-EVs for ALI in vivo. Together, these studies offer foundational mechanistic and translational insights, advancing the understanding of Treg-based therapies
Les commissions de vérité et de réconciliation à l’épreuve des perceptions de justice des victimes : réflexions critiques et pistes de solution
No full textCette recherche s’intéresse à la corrélation, généralement acceptée, mais trop rarement interrogée, selon laquelle les CVR auraient nécessairement des effets positifs pour les victimes puisqu’elles sont censées incarner une forme de justice réparatrice. Nous tentons de répondre à la question suivante : Est-ce que les CVR contribuent effectivement au sentiment de justice des victimes ? Le peu de données existantes suggère au contraire l’existence de perceptions négatives à l’égard des CVR. L’objectif de cette recherche est donc de combler ces lacunes dans la littérature, en s’intéressant à l’étude des perceptions individuelles de justice des victimes face aux CVR. Cette recherche exploratoire ne repose cependant pas sur une étude terrain, mais cherche plutôt à élaborer des pistes de réflexion qui pourront éventuellement être testées auprès des victimes.
Pour répondre à la question de recherche, nous avons mobilisé la théorie de Jason Colquitt (2001) sur la formation des perceptions de victimes. Les perceptions de justice se façonnent dans les trois dimensions de la justice : procédurale, distributive, interactionnelle, qui elle se divise en deux sous-dimensions : interpersonnelle et informationnelle. Ces trois dimensions sont distinctes, mais substituables. Par conséquent, des expériences positives ou négatives dans l’une ou l’autre de ces dimensions pourraient teinter la perception globale de la justice. En nous basant sur la littérature existante sur la justice pénale conventionnelle et la justice réparatrice, nous avons recensé les éléments potentiels dans chacune des dimensions qui peuvent influencer les perceptions des victimes vis-à-vis les CVR.
Cette grille d’analyse a ensuite été appliquée dans le cadre d’une étude de cas multiple, où les cas de la Commission de vérité et de réconciliation de l’Afrique du Sud et de la Commission dialogue, vérité et réconciliation de la Côte d’Ivoire ont été étudiés. À travers le croisement de ces deux cas, nous proposons des « interprétations contextuelles » qui nous permettent d’éclairer la théorie générale des CVR concernant les victimes. Sur la base de cet éclairage, nous proposons en dernier lieu des recommandations dans chacune des dimensions de justice pour les futures CVR, visant à maximiser les perceptions positives et à diminuer les perceptions négatives des victimes
The Effects of (A)symmetric Hand-Held Loads on Gait Performance in Young Males and Females
No full textHand-held loading is a common part of daily movement. However, its effects on gait control in young adults remain incompletely understood. Previous work suggests that arm swing, load placement, and sex may each influence gait performance, but how these factors interact during hand-held loading has not been systematically examined. Therefore, this study investigated how unilateral and bilateral hand-held loads affect spatiotemporal gait parameters, gait variability, and whole-body kinematics in healthy young adults, and whether males and females show different adjustment strategies under load. Thirty participants walked on a treadmill under three conditions: natural walking, unilateral loading at 10% body mass, and bilateral loading at 10% body mass. Whole-body 3D kinematic data were collected using a 10-camera Vicon motion-capture system, and spatiotemporal metrics and variability (CoV) were derived from heel-strike–based gait segmentation. Joint range of motion (ROM) and ROM variability were calculated for the shoulders, torso, and lower limbs.
Hand-held loading mainly influenced gait on the non-dominant side. Participants walked with shorter stride length, longer stride time, and slower stride speed when carrying a load. Shoulder swing amplitude and trunk rotation decreased across loading conditions. These changes suggest that participants relied on upper-body adjustments to maintain stability. Additionally, certain loading conditions increased gait variability, including larger left stride-time variability under bilateral loading and increased variability in shoulder and trunk rotation. Although males and females showed similar joint-angle adjustments under most conditions, some variability measures still showed sex-related differences.
Overall, this study demonstrates that hand-held loading modifies upper-body control strategies during walking and places additional demands on temporal and kinematic stability. These findings provide a foundation for future work examining how hand-held loading influences gait in older adults and clinical populations
Complementary Approaches for Improving Nasality Assessment in Cleft Palate Patients: Exploring Nasal Signal-Based Acoustic Feature and Visual Feature Analysis in Stereo Signals
No full textBackground. Cleft Palate (CP) is a common congenital condition affecting approximately 500 infants born with orofacial clefts yearly in Canada (Public Health Infobase, 2017). It disrupts the velopharyngeal sphincter, leading to oral-nasal balance disorders such as hypernasality and hyponasality. Accurate detection and classification of oral–nasal balance disorders are essential for optimizing treatment outcomes. Although auditory–perceptual judgments remain the primary method for evaluating nasality, their inherent subjectivity limits reliability. Integrating complementary measures with auditory assessments can enhance diagnostic accuracy and support better treatment planning and follow-up.
Objective. This study aimed to improve the accuracy of auditory-perceptual assessments by incorporating acoustic and visual analyses as complementary tools. The research consisted of three phases. The first study focused on exploring effective acoustic parameters extracted solely from nasal signals to develop a multiparametric approach for automatically classifying simulated oral-nasal balance disorders using linear discriminant analysis (LDA). The second study applied the developed multiparametric acoustic diagnostic algorithm to a retrospective dataset of children with cleft palate to evaluate its clinical applicability. The third study conducted a pilot investigation into the use of visual pattern inspection in stereo signals for classifying oral-nasal balance disorders.
Methods. All three studies used a retrospective, observational design based on secondary analysis of pre-existing speech data. The first and third studies employed simulated voice samples, while the second study used both simulated and clinical samples from children with cleft palate.
Results. In Study 1, LDA showed 90.9% accuracy in classifying simulated conditions based on nasal channel acoustic energy. This model demonstrated potential as a cost-effective and quantitative alternative to tools like nasometry. Study 2 confirmed that nasal-signal-derived acoustic parameters could differentiate between normal, hypernasal, and hyponasal conditions but highlighted challenges in generalizing models across datasets. Study 3 achieved an overall classification accuracy of 85.15% using visual analysis of paired oscillogram images. The highest accuracy (100%) was observed for hyponasality classification, while the lowest accuracy (61%) was recorded for mixed nasality classification.
Conclusion. This thesis advances the field of nasality assessment by demonstrating the potential of acoustic analysis of the nasal signal in isolation, as well as a new visual method for detecting and classifying oral-nasal balance disorders. These approaches could be integrated with traditional auditory-perceptual evaluations to enhance diagnostic accuracy, guide treatment planning, and support follow-up care. This research contributes new practical diagnostic approaches that could become directly relevant to clinical populations such as individuals with cleft palate or neurological disorders. The present findings lay a foundation for future work to translate these research innovations into clinical practice