14277 research outputs found
Sort by
Structure-aware communication scheduling for distributed deep learning applications
No full textWith the growing popularity of large-scale deep neural networks, efficient communication scheduling has
become crucial in distributed deep neural networks (DNNs) to reduce overall training time. In multijob
distributed training scenarios, current communication scheduling methods do not effectively utilize the
periodic communication patterns of deep learning training (DLT) jobs to reduce the potential link contention.
When multiple tenants run concurrent jobs and compete for network resources, training performance can
degrade due to increased network contention. In this thesis, we explore leveraging periodic communication
patterns to schedule DLT jobs. We analyze the performance of static shift-based scheduling strategies that
align job iterations using their least common multiple (LCM) to handle multi-job communication conflicts.
Through theoretical modeling and validation via real-world traces, we identify fundamental limitations of
such shift-based strategies, showing that about 73% of job combinations fail to benefit due to structural
constraints in job periodicity.
To address this limitation, we propose a set of structure-aware scheduling enhancements targeting both
feasible and infeasible cases. For structurally feasible combinations, we prune the scheduling search space by
leveraging the relationship between job periods and aggregation structures, and further eliminate redundant
evaluations through structural equivalence filtering. This two-step approach achieves over 99.47% runtime
reduction compared to generic Mixed-Integer Linear Programming (MILP) solvers, enabling fast and scalable
optimization. For infeasible combinations where shift-based optimization is fundamentally ineffective, we
develop a delay-tolerant strategy that reconstructs the optimization space through minimal periodic padding,
transforming unschedulable job pairs into schedulable ones. Experimental results demonstrate that our
methods significantly extend the applicability of shift-based scheduling, and can yield over 10% makespan
reduction as well as consistent fairness improvements across diverse contention scenarios
Implementing one-at-a-time therapy services as one core component of a provincial stepped care model within Prince Edward Island's community mental health and addictions services: an implementation process synthesis
No full textProviders within Health PEI's Community Mental Health and Addictions Services completed online asynchronous courses in one-at-a-time (OAAT) therapy and Stepped Care 2.0 (SC2.0) as part of an initiative to implement a provincial stepped care model. This thesis: 1) mapped the OAAT therapy implementation process using three frameworks: the Active Implementation Frameworks, Expert Recommendations for Implementing Change, and the Consolidated Framework for Implementation Research; 2) measured provider attitude and knowledge; 3) explored providers� implementation experiences; and 4) quantified OAAT therapy delivery. The study used a mixed-methods, single-cohort, observational design with two interventions. Surveys were distributed to providers at five time points over four months, and researchers interviewed seven program champions. Implementation data was abstracted from Health PEI, stakeholder, and research team documentation. Providers (N = 72) demonstrated an increase in SC2.0 knowledge and endorsed agreement for the acceptance, appropriateness, and feasibility of SC2.0/OAAT therapy, including organizational readiness (e.g., compatibility, knowledge and skills, leadership, and program champions). Interview themes aligned with existing implementation strategies (e.g., co-design, communication, and mentorship), which were considered factors for implementation sustainability. Providers delivered 3,746 OAAT therapy sessions from 2023 to 2024, showcasing the cumulative efforts made by Health PEI, providers, and stakeholders
Composting food and fishery waste: a feasibility study on system performance and microbial inoculation
No full textThis study investigates the composting processes for organic waste derived from food and fishery sources, focusing on the feasibility and performance of in-vessel (IV) and static pile (SP) systems for potential application in remote or resource-limited small communities. The research lies in its integrated evaluation of mature compost inoculation effects within an in-vessel system treating a high proportion of fishery waste (13% by weight), a feedstock combination rarely studied under controlled reactor conditions. Given the challenges posed by high moisture and nitrogen contents in such wastes, the research evaluates the effectiveness of mature compost inoculation in enhancing microbial activity, accelerating organic matter degradation, and improving compost quality. Experimental trials involving the use of IV composting reactors, with and without mature compost inoculation, and their comparison with SP composting demonstrated that the IV system with microbial inoculation sustained thermophilic conditions for seven consecutive days and achieved a 44% reduction in the carbon-to-nitrogen ratio. The germination index increased from 20% at the beginning of the process to 260% by the end, indicating excellent compost maturity. Enzyme activity analysis further supported these findings, with β-glucosidase and dehydrogenase levels in the IV system exceeding those in the SP system by 32% and 26%, respectively. Acidic and alkaline phosphatase activities were also consistently higher in the IV system. These advantages highlight the superior decomposition rate, pathogen inactivation potential, and nutrient stabilization offered by the IV system compared to traditional SP composting. Additionally, the study evaluated the operational performance of an IV composting reactor equipped with an axial mixer, demonstrating its potential for efficient waste treatment in remote or resource-limited settings. The
findings provide practical applications for decentralized organic waste management in small northern or rural communities and the industrial, commercial, and institutional sectors, enabling environmentally friendly waste diversion, reduced landfill dependency, and production of high-quality compost for agricultural or land restoration use. Overall, the results confirm that combining controlled composting conditions with microbial inoculation significantly enhances process efficiency, compost stability, and product quality, supporting the adoption of IV systems as a sustainable solution for managing organic waste generated by the Industrial, Commercial, and Institutional sector and in small and remote communities
Mapping the entrepreneurial ecosystem of the cultural and creative industries: an examination of the Corner Brook region, NL, Canada
No full textThis thesis examines the application of the Entrepreneurial Ecosystem Mapping (EEM) framework developed by Stam and Van de Ven (2021), to the Cultural and Creative Industries (CCIs) in Corner Brook, Newfoundland and Labrador, Canada. The research aims to assess the current state of the area's CCIs entrepreneurial ecosystem, evaluate its sustainability, and identify opportunities for fostering creative entrepreneurship in this small, resource-constrained region. By combining secondary data analysis with insights from the researcher's embedded experience, the study provides a comprehensive understanding of the ecosystem's elements, interactions, and challenges.
The findings demonstrate that Corner Brook's CCIs ecosystem is emerging with significant strengths, including robust cultural and natural assets, educational infrastructure, and local champions for creative initiatives. However, challenges like limited CCI-specific policies, funding gaps, talent retention issues, and lack of data, hinder its growth. The study proposes an adaptation to Stam and Van de Ven's (2021) EEM framework to better reflect the dual cultural-economic nature of CCIs, and to specifically integrate cultural value, natural capital, and tourism as key elements of the CCIs entrepreneurial ecosystem (Throsby, 2000).
This research contributes to entrepreneurial ecosystem theory by demonstrating not only the adaptability of the EEM framework to CCIs and small, less urban regions, but also by proposing an adaptation of the model based on the findings and specifically tailored for CCIs. It also provides actionable insights for policymakers and stakeholders, highlighting the importance of customized strategies to support creative entrepreneurship and foster sustainable regional development
Estimating whole-body tissue composition from sub-body CT scans using automated field-of-view detection
No full textBody composition is a critical health indicator with profound implications for clinical
conditions, including sarcopenia, cancer, cardiovascular diseases, osteoporosis, and
diabetes. Accurate assessment of body composition is essential for personalized and
preventive medicine, as it aids in clinical decision-making and health optimization.
This thesis investigates the prediction of whole-body tissue composition from subbody
CT scans, focusing on four key tissue types: skeletal muscle (SKM), subcutaneous
adipose tissue (SAT), visceral adipose tissue (VAT), and intermuscular adipose
tissue (IMAT).
The research has been conducted in two phases. The first phase involves developing
a deep learning model to classify sub-body CT scans into six anatomical regions:
chest (CHE), abdomen (ABD), pelvis (PLV), chest & abdomen (CHA), abdomen &
pelvis (ABP), and chest, abdomen & pelvis (CAP). Utilizing maximum intensity projection
to map 3D scans to 2D, the CNN-based model achieved over 95% classification
accuracy across diverse patient demographics and clinical protocols.
The second phase focuses on developing regression models to predict whole-body
tissue volumes using features derived from classified sub-body scans. For each classified CT scan, a separate multivariate linear regression model is developed, incorporating
demographic and CT-derived features. These models were validated using ten-fold
cross-validation, achieving high performance across regions. For individual regions,
SKM demonstrates strong performance across all areas with high R² values (≈ 0:9)
and low MAD (e.g., chest: 1.09 ± 0.86 L). SAT performs best in the PLV region (R²
= 0.894, MAD = 1.59 ± 1.29 L), while VAT achieves its highest performance in the
ABD region (R² = 0.965, MAD = 0.26 ± 0.21 L). IMAT shows small MAD values
(e.g., chest: 0.26 ± 0.20 L) but relatively low R², with slightly better predictability in
the CHE (R² = 0.790). For combined regions, SKM performs best in the CAP region
(R² = 0.949, MAD = 0.77 ± 0.65 L), while SAT improves in the ABP region (R² =
0.927, MAD = 1.36 ± 1.13 L). VAT shows consistently excellent performance across
all combined regions, with similar R² values (CHA: 0.973, ABP: 0.979, CAP: 0.999)
and low MAD. The broader anatomical coverage in combined regions enhances predictability,
particularly for SKM, SAT, and VAT, as it captures more comprehensive
tissue distribution.
These findings highlight the feasibility of using sub-body CT scans to estimate
whole-body tissue composition accurately, reducing the need for resource-intensive
full-body scans while minimizing radiation exposure. This approach holds promise
for advancing personalized healthcare strategies and preventive interventions.Includes bibliographical references (pages 58-68
Investigating the presence of abandoned, lost, and discarded fishing gear (ALDFG) to protect golden cod in the Gilbert Bay MPA
No full textAbandoned, lost, and discarded fishing gear (ALDFG) is a global environmental, economic, and social issue. This thesis examines the magnitude of ALDFG in Gilbert Bay, Labrador to protect the most genetically distinct population of Atlantic cod (Gadus morhua) in the Western Atlantic, whose numbers have declined since creation of the Marine Protected Area (MPA) in 2005. Expanding on community-led initiatives, key knowledge holders (n = 14) were interviewed to obtain qualitative and geospatial data to guide the understanding and investigation of ALDFG. According to knowledge holders, ALDFG was not believed to be an issue impacting the Gilbert Bay cod, primarily due to commercial scallop fishers incidentally dragging up lost gear. During retrieval, a total of 66 sea-based sites were investigated, yielding a single cod trap. Knowledge holders also noted land-based gear in derelict stages and on wharves as an area of concern. As a result, 18 land-based sites were identified, retrieving various amounts of fishing nets, trawl lines, crab pots, cod traps, and fishing rope from 10 different sites between July and November of 2021. Data collected from the interviews, literature review, and field work are also used to provide further policy recommendations to ALDFG and fisheries management for Gilbert Bay and the region overall
Design and analysis of a hybrid powered reverse osmosis water system for use in a remote location in Newfoundland
No full textRemote communities such as McCallum, Newfoundland and Labrador, face critical challenges in accessing clean water and reliable electricity. Persistent water shortages, combined with lead contamination from naturally occurring soil, have rendered conventional solutions ineffective. Additionally, the community lacks grid access and depends entirely on diesel generators, which are expensive, emission-intensive, and often unreliable. To address these issues, this thesis presents the design, simulation, and implementation of a hybrid-powered reverse osmosis (RO) water treatment system.
The work is structured in three phases. First, an optimal hybrid energy system (HES) was developed using HOMER Pro software, combining photovoltaic panels, a wind turbine, battery storage, and a small DC diesel generator for backup. This system reduced net present cost by over 70% compared to diesel-only operation and achieved a 98.8% renewable energy fraction, significantly cutting greenhouse gas (GHG) emissions. Second, dynamic simulations in MATLAB/Simulink validated system stability and reliable power delivery to the RO unit under varying environmental conditions. Third, a low-cost and low-power SCADA system was implemented using LoRa-enabled ESP32 modules for long-range communication and a local MQTT broker with web-based FUXA for visualization and monitoring. This architecture supports real-time monitoring and two-way control without relying on internet or cellular connectivity.
The complete system was tested in the lab under various operational scenarios, confirming its ability to deliver clean, reliable, and low-emission power. This thesis offers a scalable and replicable model for resilient energy infrastructure in off-grid, resource-constrained communities
Wetland characterization through multi-sensor satellite data and advanced AI models
No full textWetlands are generally defined as areas that are inundated or saturated with water
for at least part of the year, though definitions can vary significantly across different
scientific disciplines. Recognized as some of Earth's most valuable and productive
ecosystems, wetlands provide a range of essential ecological services. These include
regulating global climate through carbon storage, naturally purifying water by filtering
pollutants, mitigating
ood and drought impacts, protecting shorelines from
erosion, conserving soil, and supporting biodiversity by serving as critical habitats
for diverse wildlife species. Wetlands also offer spaces for recreation, cultural appreciation,
and aesthetic enjoyment, underscoring their multifunctional importance for
environmental health and human well-being. They are dynamic ecosystems characterized
by complex and
uctuating vegetation, water, and soil interactions. Traditional
monitoring methods rely on in-situ sampling of vegetation, hydrology, and soil, which
are labor-intensive, time-consuming, and challenging to scale across vast or inaccessible
regions. In contrast, remote sensing (RS) technology, coupled with advanced
artificial intelligence (AI) techniques, offers an efficient, scalable, and non-intrusive
approach for assessing wetland health, spatial extent, and ecological status. Through
the integration of multisource - multispectral, Synthetic Aperture Radar (SAR), and
Light Detection and Ranging (LiDAR) data - RS enables high-resolution monitoring
across long-term periods and expansive spatial scales, allowing for efficient mapping
and characterization of different wetland classes. These tools are crucial in quantifying
wetland changes, identifying anthropogenic impacts, and understanding seasonal
ecological variations in wetlands, thereby facilitating data-driven conservation and
sustainable management efforts.
This thesis focuses on improving wetland mapping and characterization by introducing
novel methodologies that utilize advanced RS technologies and multisource
satellite data. First, it presents a comprehensive literature review on wetland monitoring,
focusing on the integration of RS and machine learning (ML) techniques. In
the context of rapidly changing wetland landscapes, the synergy between multisource
RS data and ML algorithms offers an opportunity to enhance decision-making support
for wetland management. The review provides guidance on selecting suitable
ML algorithms and RS data types for detailed wetland monitoring. A second key
contribution of this work is the development of a novel deep learning (DL) framework
to address the challenge of limited sample data, advancing the potential of RS techniques
for wetland mapping. Emphasizing the integration of SAR and optical data,
the framework utilizes multimodal data sources and applies Graph Convolutional Networks
(GCN) and Convolutional Neural Networks (CNN). The experimental results
demonstrate significantly improved classification accuracy, particularly in mapping
complex ecosystems, compared to single-stream approaches. Thirdly, this research
highlights the importance of RADARSAT Constellation Mission (RCM) compact polarimetric
SAR (CP-SAR) data in wetland studies. A novel framework is developed
to extract robust features for wetland mapping, showcasing notable improvements in
wetland classification performance compared to existing models. Fourthly, a new SAR
decomposition technique is introduced, designed specifically for CP-SAR data from
RCM imagery. This technique refines the analysis of wetland structures and compositions.
Additionally, a unique descriptor termed compact polarimetry signature
(CPS) is proposed based on time-series RCM data, enabling a detailed investigation
of wetland phenological cycles and associated SAR backscatter variations during
active growing seasons. The results demonstrate the superior effectiveness of these
techniques in characterizing complex wetland landscapes, surpassing the accuracy and
detail of traditional single-date approaches. Finally, this work evaluates the added
value of incorporating diverse data sources, including optical, SAR, and LiDAR data,
into wetland mapping. Using the Google Earth Engine (GEE) platform, two main
objectives are pursued: (1) integrating Global Ecosystem Dynamics Investigation
(GEDI) LiDAR footprint heights with multisource datasets to generate vegetation
canopy height (VCH) maps, and (2) enhancing wetland mapping by using VCH as a
predictive input. Results highlight the critical role of VCH derived from GEDI samples
in improving wetland classification accuracy, offering a vertical vegetation profile
that enriches understanding of wetland ecosystems. This thesis introduces innovative
methodologies that pave the way for practical and scalable solutions in wetland mapping.
The approaches outlined not only enhance our ability to map complex wetland
systems in Canada but also provide adaptable strategies applicable to ecologically
similar wetlands around the world.Includes bibliographical reference
Moho structure of northeastern Canada using receiver function analysis
No full textThis study investigates variations in Mohorovi£i¢discontinuity (Moho) depth across the northeastern
segment of the Canadian Shield using the receiver function (RF) technique. The study
area encompasses a range of geologically distinct terranes, including both Archean and Proterozoic
domains, making it an ideal setting for investigating the evolution of the continental crust from the
Paleoarchean era to the emergence of plate tectonic activity in the Paleoproterozoic. Central to
this investigation is the Trans-Hudson Orogen (THO) — a Paleoproterozoic suture zone situated
between the Superior Craton in the southeast and the Rae and Hearne Cratons in the northwest.
This region retains a complex geological history that offers key insights into processes of crustal
growth and continental assembly.
We utilize teleseismic earthquakes with magnitudes (M > 5.5) recorded by both permanent and
temporary seismic stations to compute Receiver Functions (RF) and estimate Moho depth beneath
each station. While the standard RF method remains central to this work, we compare our findings
with previous studies and models (e.g., Crust 1.0) to validate our interpretations. Although we
attempted to assess azimuthal variation in Moho depth, our efforts were limited by the uneven
azimuthal distribution of incoming seismic waves. This restricted our ability to reliably separate
directional effects such as dipping interfaces or seismic anisotropy, particularly in geologically
complex areas like the Trans-Hudson Orogen. Despite the methodological strengths, our analysis
is constrained mainly by the limited azimuthal distribution of seismic events, which hampers our
ability to robustly resolve directional variations in Moho depth and distinguish between seismic
anisotropy and structural dips.
Moho depth estimates in the Superior Province (Archean) range from 34 to 40 km, reflecting relatively
simple crustal structure. In contrast, the THO exhibits more variation, with Moho depths
between 35 and 43 km. While some of our stations reveal sharp receiver function peaks indicative
of well-defined Moho boundaries, others display more complex or ambiguous patterns—likely
influenced by uneven data coverage, crustal anisotropy, topographic variations in the Moho, or
differences in its nature, ranging from sharp to gradational transitions.
Regional patterns also emerge from our analysis. In Greenland, Moho depth increases from
north to south, while in the Canadian Shield, more spatially variable trends are evident.
Overall, this work demonstrates the value of receiver functions in probing crustal structure across
a geologically complex region, while also emphasizing the importance of dense seismic coverage
and integrated methodologies. Future studies would benefit from the inclusion of complementary
techniques (e.g., joint inversion, gravity or magnetotelluric (MT) surveys) to better resolve
anisotropic features and refine Moho depth estimates in areas with limited seismic data
Examination of seismic models: inhomogeneity, anisotropy, and Backus averaging
No full textThe thesis examines seismic models of the Earth in terms of inhomogeneity, anisotropy, and Backus averaging. Chapter One provides background information. Chapters two to four are described below. The fifth chapter provides concluding remarks. The common thread through all the chapters is the use of data acquired in the same borehole: Vertical Seismic Profiling (VSP) data and a sonic log.
A study on the estimation of inhomogeneity and anisotropy parameters from walkaway VSP traveltime data, using a multi-layered mathematical model, is presented in the second chapter. Least-squares residuals between measured and modelled traveltimes are minimized to estimate the anisotropy parameter, χ, and inhomogeneity parameters, a and b, of the layers. A two-step optimization is performed, and an adaptation of the Nelder-Mead algorithm is used to estimate the parameters. The methodology is applied to synthetic data and then to real data. An assessment of the reliability of results subject to noise shows the noise threshold to be quite low. Beyond this threshold, parameter estimates have diminishing accuracy. Using synthetic data, parameters are reliably estimated. With real data, parameter estimations indicate anisotropy to be exhibited only in the bottom layer.
The third chapter is on the use of the Bayesian Information Criterion (BIC) for the selection of a model that is most representative and has the fewest number of parameters to fit the data. Eight three-layer models, with different parameterizations, are considered that correspond to the medium in which the VSP data were acquired. The simplest model
is inhomogeneous and isotropic with six parameters. The most complicated model is inhomogeneous and anisotropic and consists of nine parameters. BIC values indicate the best model as the one with seven parameters and anisotropy in the third layer.
An adaptation of the Backus average to obtain more accurate traveltimes for obliquely propagating waves is presented in the fourth chapter. A weighting is applied that considers the distance travelled in each layer, with weights corresponding to source-receiver offsets. Traveltimes computed from the standard Backus average are compared to traveltimes computed using the modified Backus average in three cases. The first, a ten-layer synthetic model with a 30-degree take-off angle, the second, with an extreme distance of 7000 m, and the third, with real data. In all three cases, the modified Backus average performs better.
All three objectives: estimation of inhomogeneity and anisotropy parameters, use of BIC to indicate the most representative model of the medium based on the fit of the data, and modification of the Backus average to correct for non-vertical raypaths to obtain more accurate traveltimes, were successfully achieved.Includes bibliographical references (pages 130-137