82837 research outputs found
Sort by
Wellbeing in assessment network analysis
No full textThis the first of a series of Research Briefs created from the Intrinsically Motivating Assessment Practices (iMAP) project. It is from data collected in Fall 2024
Automated and Efficient Deep Neural Network Design via Quantifiable Data Science
No full textDeep Neural Networks (DNN) are a major driving force behind the tremendous progress in real-world tasks that Machine Learning (ML) and Artificial Intelligence (AI) have made. However, the computational resources required by DNNs limit their deployment in various scenarios. Thus, there is a need not only to develop DNNs that not only achieve superior performance, but are efficient in terms of hardware-friendliness metrics like latency and model size. This objective presents two key design challenges: Constructing more efficient neural networks and finding intelligent ways to compress existing ones.
Neural Architecture Search (NAS) solves the former by automating the discovery of optimal DNNs. Performance predictors lower the compute requirement of NAS but are limited by the constraints of predefined search spaces and benchmark tasks. To address this, we propose GENNAPE and AIO-P. GENNAPE introduces a robust Computational Graph format that can represent architectures from different search spaces. AIO-P utilizes knowledge injection techniques to expand the scope of predictors into complex real-world tasks with limited data, like panoptic segmentation. Further, we propose AutoGO which constructs a database of architecture components to optimize the primitive operation structure of existing architectures. This framework maximizes performance and hardware-friendliness in deployment scenarios. In the zero-shot transfer setting or with minimal fine-tuning, GENNAPE achieves over 0.85 SRCC on several search spaces, while AIO-P outperforms several zero-shot proxy methods at cross-task performance prediction. Further, AutoGO improves the design of the best architectures from several public NAS-Benchmarks.
On the other hand, model compression reduces the hardware costs of existing DNNs, e.g., quantization lowers model bit precision. However, effective compression should identify which components of a model are crucial to maintaining performance, while existing techniques aim to minimize end-to-end performance loss. To address this, we propose an data science-based approach to extract interpretable insights about the block choices that compose DNNs and how they behave on different hardware devices. We further propose AutoBuild, a data-efficient method for constructing high performance and hardware-friendly DNN architectures. Further, we propose QuaSeDiMo to identify the individual weight layers and operations in Diffusion Models that are sensitive to low-bit quantization. In practice, AutoBuild allows us to find reduced Stable Diffusion v1.4 Inpainting variants with less than 100 labeled samples, while QuaSeDiMo finds effective sub 4-bit quantization configurations for a myriad of text-to-image models that outperform other approaches in terms of CLIP and FID metrics while generating superior visual content in terms of human preference
A Tracheobronchial Specific in vitro Dissolution Test – A Problem Well Filtered is a Problem Half Dissolved
No full textInhaled pharmaceutical aerosols have been widely successful in treating a variety of lung diseases
by delivering drugs directly to the site of action. For any therapeutic action to occur, solid drug
particles must deposit on the airway surface, dissolve into lung fluid, and be absorbed by the local
tissue. Accordingly, the efficacy of inhaled drugs can be improved by ensuring the drug is delivered
to the targeted region in which its designed to treat when a local effect is desired. To better
understand the dissolution step, the pharmaceutical industry relies on dissolution tests to evaluate
the release of drug products delivered in solid dosage forms. For orally inhaled drug products, the
development of dissolution tests is still an active field of research, owing in part to the challenge of
collecting aerosol delivered from inhalers in quantities representative of those delivered to the
targeted lung regions, in a manner suitable for presentation to a dissolution test apparatus. Although
many dissolution test methods for inhaled drugs have been developed and described in literature,
none currently accounts for regional lung deposition - despite its significant impact on therapeutic
outcome.
The aim of this thesis was to develop an in vitro dissolution test for inhaled drugs depositing in the
tracheobronchial lung region. A recently developed in vitro apparatus for measuring regional lung
deposition, which uses stainless-steel filters to capture the tracheobronchial dose, was used as the
drug collection method. The design criteria for the dissolution test apparatus were: (i) to be
insensitive to loaded dose (i.e., produce consistent results regardless of captured drug mass), and
(ii) to be able to produce discriminatory dissolution results between inhalers with known dissolution
differences. Prior to any dissolution experiments, filters were loaded with increasing drug mass using
the deposition apparatus to first assess filtration sensitivity to loaded dose, revealing a linear
relationship between delivered dose and captured drug mass.
The custom dissolution test apparatus consisted of the tracheobronchial filters with captured drug
being dissolved in a phosphate buffered saline containing 0.5% w/v sodium dodecyl sulphate
solution at 37 °C. Samples were withdrawn at specific time points and quantified using highperformance liquid chromatography. The dissolution characteristics of the tracheobronchial dose of
a commercial pressurized metered-dose inhaler and dry powder inhaler were compared. Dissolution
measurements revealed: (i) no significant differences in dissolution results with increased drug
mass, indicating that the method was insensitive to loaded dose, and (ii) slower dissolution rates for
the pMDI as compared to the DPI, consistent with published findings.
Collectively, these findings demonstrate the developed method’s robustness, repeatability, and
ability to differentiate dissolution characteristics between formulations. The work done in this thesis
could be applied in the testing of orally inhaled drug products, for quality assurance/control,
evaluation of generics, drug development, and contribute to improved understanding of the
dissolution of inhaled pharmaceutical aerosols in the lung fluids
Periphyton communities as bioindicators of environmental changes in lotic ecosystems
No full textBiofilms containing freshwater algae attached to submerged surfaces, termed periphyton, can serve as indicators of the effects of multiple environmental changes over time while also being a critical link between the abiotic environment and biota. Freshwater ecosystems are among the most vulnerable to anthropogenic loss of biodiversity and impaired ecosystem function. Often, a combination of novel or extreme environmental changes attributable to human activities are collectively the root cause of declines in biodiversity and ecological function in lotic ecosystems (e.g., streams and rivers). Such “stressors” frequently interact with each other in space and time to result in unexpected cumulative impacts, or “ecological surprises”. Uncertainties over the nature of these interactions are at the core of the knowledge gap that exists in the emerging field of multiple stressors ecology.
A survey of periphytic algal communities was first conducted along the Bow River in Banff National Park, Canada, to evaluate the potential anthropogenic impacts of the Town of Banff and its wastewater treatment plant on surface water quality. High performance liquid chromatography of chlorophyll a (i.e., a surrogate measure of algal biomass) and taxonomically diagnostic pigments revealed that chlorophyll a concentration and communities in sites downstream of the townsite and wastewater treatment plant were significantly distinct from upstream references. Cyanobacterial pigments were significantly more concentrated downstream of the townsite and the wastewater treatment plant. Specific conductivity and total phosphorus were the primary abiotic factors explaining the taxonomic variance among algal communities within these reaches. Although all three algal groups (chromophycean, chlorophycean, and cyanophycean) were more associated with the sites downstream, their concentrations there were not high enough to warrant management concern. These findings establish a baseline for future environmental assessments of the Bow River Basin.
Next, a larger survey of algal communities was conducted along tributaries of the North Saskatchewan River across Alberta to assess the effects of local and regional factors on community structure. Structural equation modelling (SEM) revealed that regional land-use variables were rarely linked directly to either the periphyton or phytoplankton communities. However, several strong terrestrial-aquatic linkages were identified between land-use and in-stream physicochemical variables, namely dissolved organic carbon (DOC), total phosphorus (TP) and dissolved inorganic nitrogen (DIN). In turn, landscape features, such as agricultural and residential area, indirectly explained spatial variance in the algal communities across the study area through their influence on abiotic stream environments. The inferred indirect effects of land use activities on periphyton communities were more pronounced and complex than seen in phytoplankton communities. Thus, these findings suggest periphyton are likely better bioindicators than phytoplankton of the cumulative impacts of multiple human land-use stressors within streams.
To further explore the potential direct and interactive effects of agricultural practices on freshwater periphyton, I conducted a factorial experiment exposing stream mesocosms to two commonly used agrochemicals, glyphosate and bromoxynil. Contrary to my original prediction, the combination of both herbicides, when applied simultaneously, had a stimulatory effect on the algal community biomass. When applied separately, glyphosate and bromoxynil had deleterious effects on algal diversity and abundance, except for chromophytes. As both herbicides release nutrients upon degradation, the positive effects on algal growth in the combined treatment could be explained by the release from co-limitation by nitrogen and phosphorus. The unexpected results the combination of these herbicides had on algal growth emphasizes the importance of studying the cumulative effects agrochemicals have on aquatic systems.
Overall, I used a comparative experimental approach showing how periphyton communities are useful as bioindicators of environmental change in lotic ecosystems. Several lines of evidence (e.g., chlorophyll, taxonomically diagnostic algal pigments, genus-level data, and ecosystem function) were documented to provide comprehensive assessments of periphyton responses to a variety of natural disturbances and anthropogenic stressors across a broad spatial scale (i.e. biogeographic to in vitro). My research builds on the rather extensive field of bioindicators, highlighting the use of pigment analysis as a unique and a more cost and time-effective method, when compared to the more traditional algal taxonomic approach. As little is documented about periphyton communities in Alberta, this research provides baselines for future assessments of lotic ecosystems within the province
Design, Techno-Economic Assessment, and Comparison of Clean Hydrogen Production/Utilization Technologies Based on High-Temperature Solar Thermal Energy
No full textThe transition to a sustainable and low-carbon energy future significantly depends on the development of economically viable and environmentally sustainable hydrogen production pathways. Achieving an optimal balance between economic feasibility and environmental impact remains a formidable challenge within the hydrogen sector, particularly in large-scale applications. Concentrated solar power (CSP), with its superior efficiency over other renewable sources, offers immense yet almost underexplored potential for high-temperature, solar-driven hydrogen production. This thesis is dedicated to exploring the integration of CSP in water electrolysis and methane pyrolysis, advancing novel design proposals to enhance these solar-powered processes. Through a comprehensive comparative analysis, the research highlights each method’s unique advantages and constraints, revealing their roles in fostering a resilient hydrogen economy. Moreover, the thesis investigates the potential of hydrogen storage and hydrogen-based combined heat and power (CHP) systems for sustainable energy management, with a focused analysis on their implementation across the University of Alberta, Alberta, and Canada. This work aims to lay a foundation for next-generation solar-driven hydrogen technologies and integrated energy solutions, contributing to the broader decarbonization and sustainability goals.
First, a comprehensive assessment is presented, including innovative designs and evaluation of high-temperature solar thermal field integration with proton exchange membrane electrolyzers (PEME), solid oxide electrolyzer cells (SOEC), and methane pyrolysis. Through comparative evaluation of low- and high-temperature water electrolysis technologies, it was determined that the SOEC integration with high-temperature solar thermal field yields a high exergy efficiency of 74%, with payback periods of less than two years in reversible operation. However, despite selecting an optimal working fluid to enhance the operational efficiency of the solar unit and conducting a thorough techno-economic optimization, the exergy efficiency of the CSP-PEM configuration remained constrained to approximately 18%, with an associated cost rate of 1.93/kg for the proposed novel design, which employs molten salt as a thermal transfer medium between high-temperature solar thermal fields and methane pyrolysis. An exploration of hydrogen incentives in the U.S. and Canada for clean hydrogen production indicates that the LCOH could be reduced to as low as 2.83/kg, outperforming SSOEC’s LCOH of 1/kg, highlighting the need for government support mechanisms, such as clean hydrogen tax credits, to enhance the economic feasibility of SSOEC systems.
In addition, a comparative case study examines SOFC-based CHP and hydrogen internal combustion engine (HICE-based CHP) systems using operational data from the University of Alberta. Despite higher capital and operational costs for SOFC-based CHP, its potential for superior long-term performance and sustainability exceeds that of HICE-CHP, which is limited by direct nitrogen oxides (NOx) emissions. Finally, this thesis evaluates the potential of SOFC-CHP systems for Alberta and Canada, forecasting market impact, job creation, and greenhouse gas (GHG) reduction potential. Results indicate significant benefits, with projected GHG reductions of 7.8 million tons in Canada’s commercial sector and 1.7 million tons in Alberta by 2050, alongside substantial job creation and a potential Canadian market size reaching $15 billion. This thesis thus positions SOFC-based CHP as the favorable choice for sustainable energy management as the technology matures, underscoring the potential of high-temperature solar thermal integration and SOFC technology for a resilient hydrogen economy