82837 research outputs found
Sort by
Exploring AI, Power, and the Planet Through Literature: An AI Literacy Proposal for a High School ELA Classroom
No full textAbstract
This literature review, conducted as part of a Roger S. Smith Undergraduate Student Research Award under the supervision of Dr. Claudia Eppert, examines scholarship at the intersections of artificial intelligence (AI), pedagogy, and environmental concerns, with a focus on English Language Arts (ELA) education. As AI becomes more prevalent in classrooms, research highlights both instructional benefits, such as efficiency and accessibility, and significant risks, including academic integrity, equity, privacy, and environmental impacts.
The review synthesizes three strands of literature: AI in education, AI in ELA and the arts, and AI’s relationship to climate change. Findings indicate cautious optimism among educators and students, alongside concerns about reduced critical engagement and originality, positioning AI as an emerging literacy. Literary and speculative texts are identified as effective pedagogical tools for exploring ethical, technological, and ecological questions. Research on AI and climate change further underscores the substantial resource demands of AI systems and their potential role in climate education
Sticks and Stones: Discourses of Wellbeing and Overwhelm among Staff at a Youth-Serving Agency
No full textHow did Facebook influence the outcome of Zambia’s 2021 elections?
No full textAbstract: The Zambian electoral process has undergone constant transformation. As a growing democracy, the country held its last general elections in 2021, whose outcome favoured the opposition. Following the loss of the election by the ruling Patriotic Front (PF) by over a million votes to the opposition Hakainde Hichilema-led United Party for National Development (UPND), its party president Edgar Lungu resigned from the party position and announced his retirement from active politics, despite calls from his supporters to petition the outcome. My research focuses on investigating the impact of Facebook usage on the result of the 2021 Zambian election, based on the research question: how did Facebook users influence the outcome of the 2021 Zambian elections? Using document analysis, a qualitative research method, the study examines how online interactions on Facebook influence public opinion, voter participation, and political outcomes, drawing on public documents, opinion poll results, media reports, Facebook content related to elections, and academic research journals. The study focuses on the role of Facebook users in shaping the electoral narrative, influencing voter behaviour, and mobilizing support for political candidates. The results of this research offer valuable insights into how social media is transforming the dynamics of electoral campaigns in emerging democracies and provide lessons for future electoral processes. This research is significant for policymakers, political actors, and scholars interested in the intersection between technology and politics
Evaluating Methodologies used to Measure Carbon Nanomaterials in Occupational Settings
No full textCarbon nanomaterials (CNMs) have valuable properties, including high mechanical strength, flexibility, exceptional electronic capabilities and high thermal conductivity. As a result, they are increasingly integrated into a variety of industrial applications. However, this raises concerns for workers exposed to CNMs during their synthesis, handling or integration into products. Furthermore, despite the known toxicity of engineered nanomaterials, there is currently no standardized methodology to assess CNM exposure in occupational settings. Therefore, the work presented in this thesis aims to evaluate current sampling and measurement procedures in assessing exposure to CNM in occupational settings. This thesis is composed of four chapters, including an introduction (Chapter One), a review of the literature (Chapter Two), original research (Chapter Three) and a general conclusion (Chapter Four). In Chapter Two, we evaluated key methodologies currently used to characterize exposure to CNMs by reviewing the body of evidence. We conducted a comprehensive search on major databases and Google Scholar from inception through March 2024. Out of 172, 37 studies were included in the review. This review synthesizes the various methods into one document and can serve as a resource for selecting appropriate instruments and methods based on the occupational environment.
Chapter Three includes the results of a research study aimed at assessing a practical method for collecting and analyzing CNMs in workplace environments. An exposure chamber was used to evaluate the sampling performance of three respirable samplers and two cascade impactors when collecting five types of aerosolized CNMs. The respirable samplers included an Institute of Occupational Medicine (IOM) sampler with polyurethane foam, a cassette with a cyclone assembly and a parallel particle impactor (PPI). The cascade impactors tested were the Mini-MOUDI eight-stage impactor and the Sioutas five-stage impactor. The tested CNMs included multi-walled carbon nanotubes (MWCNT), single-walled carbon nanotubes (SWCNT), carbon black (CB), carbon nanofibers (CNF) and graphene. Gravimetric measurements showed consistent results for MWCNT, SWCNT, CB, and CNF across multiple runs, with errors ranging from 4% to 27% for all samplers except the Mini-MOUDI. The weights from the Mini-MOUDI sampler were repeatedly below the detection limit for the MWCNT, SWCNT and CNF experimental runs. Results also revealed a significant difference in the mass concentration collected between the cyclone and the Mini-MOUDI for carbon black (p-value<0.0001). Additionally, when measuring SWCNT, the percentage of elemental carbon collected by the IOM sampler was significantly higher than cyclone (p-value<0.05). Further analysis involved Raman spectroscopy on the collected filters to identify important spectral features of each CNM, including peak positions and the ratios between different peak intensities. A 50:50 mixture of MWCNT and CB was also sampled, and Raman spectroscopy demonstrated that it is possible to differentiate between the mixture and the individual CNMs. Finally, scanning electron microscopy provided insight into particle morphology and aggregation states. This study's results proposed the most effective samplers and offline methods for assessing exposure to CNMs in occupational settings. These findings can be valuable in protecting workers from potential health risks
Multistep Credit Assignment in Deep Reinforcement Learning
No full textTemporal credit assignment in reinforcement learning (RL) is the problem of associating rewards with the past actions that contributed to them. Temporal-difference (TD) learning assigns credit by comparing the observed and expected outcomes after each action, generating a TD-error signal which drives the improvement of future predictions. This incremental, 1-step learning approach eventually grasps long-term consequences but requires many repetitions to converge. In contrast, multistep approaches can learn much faster by reapplying TD errors across multiple actions.
Eligibility traces were the main way to implement multistep credit assignment in RL, by broadcasting errors over a fading record of recent observations. However, over the past decade, RL algorithms have largely adopted neural networks to approximate value functions. These deep RL methods depend on randomized experience replay to stabilize learning, precluding the use of eligibility traces.
As a result, advanced multistep credit-assignment techniques have not been widely adopted in deep RL. The vast majority of existing replay methods opt for n-step returns, which are computationally cheap but nonsmooth. On the other hand, λ-returns offer a potential surrogate for eligibility traces, but they are expensive and not commonly used in large-buffer RL agents. In both cases, off-policy corrections are typically ignored, incurring bias in the return estimation.
In this thesis, I establish new properties of compound returns—especially λ-returns—and discuss how to leverage them efficiently in deep RL to accelerate learning. I prove that compound returns generally improve the bias-variance trade-off compared to n-step returns. Additionally, I show that all compound returns obey the recency heuristic from psychology: the influence of a reinforcing event decreases monotonically in elapsed time. Long-tailed compound returns like λ-returns, however, are most effective for credit assignment in practice. I propose a cached-trajectory replay strategy for efficiently computing λ-returns while mitigating truncation bias and sample correlations. Finally, I generalize compound returns with trajectory-aware returns, which jointly consider experiences in the replay memory to produce more efficient off-policy corrections. Throughout this thesis, I evaluate the proposed algorithms in a variety of tabular environments, imaged-based games, and robot simulations
Unfolding Structure from Sequence: A Computational Perspective on Genomic and Molecular Organization
No full textThis thesis presents a comprehensive investigation into promoter sequence similarity patterns across the human genome and their implications for gene regulation and 3D chromatin structure, integrating the development and application of a novel sequence comparison tool, SEQSIM (Sequence Similarity). The thesis contains three core chapters, each addressing complementary aspects of promoter sequence similarity and organization: the development of SEQSIM and its benchmarking with CABS1 (calcium binding spermatid associated 1); a focused case study on the PRAMEF (Preferentially Expressed Antigen in Melanoma Family) genomic neighborhood; and a genome-wide analysis of promoter similarity patterns, their frequency, distribution, and structural correlates.
We first introduce SEQSIM, a global alignment-based tool designed to compare upstream promoter sequences of human protein-coding genes. SEQSIM employs a custom-made algorithm to detect regions of sequence similarity in the 2k nucleotide (nt) upstream promoter space, enabling fast, pairwise comparisons across all human genes. Unlike traditional local alignment tools or clustering approaches, SEQSIM provides an unbiased method to quantify similarity without assumptions of functional conservation or transcription factor binding site reuse. Through benchmarking, SEQSIM demonstrates high sensitivity in detecting both paralogous promoter homology and unexpected similarity patterns between loci otherwise thought to be unrelated. Potentially, the tool offers intuitive outputs that can be integrated with 3D chromatin conformation datasets, epigenomic annotations, and gene expression data.
Building upon SEQSIM’s analytic capabilities, the next chapter applies the tool to investigate a genomic region on chromosome 1 containing the PRAMEF, HNRNPCL, LINC and RNU6 gene families. These gene clusters are characterized by high sequence redundancy. SEQSIM reveals dense intrachromosomal promoter similarity among PRAMEF family members and identifies similarities in adjacent non-PRAMEF genes as well. These observations suggest that the PRAMEF locus constitutes a unique regulatory neighborhood shaped by repeat element groupings, and potentially higher-order chromatin folding.
The following chapter extends the application of SEQSIM to a genome-wide scale to identify and categorize promoter similarity patterns throughout the entire human genome. The analysis identifies 79 discrete promoter similarity clusters, many of which are distributed across distant chromosomal locations. The results propose that promoter-level sequence similarity may contribute to long-range chromatin organization, potentially acting as a latent code guiding regulatory proximity. Furthermore, the sequence features of high-similarity promoters suggest that sequence composition itself may facilitate architectural recognition or enhancer-promoter compatibility.
Future studies should aim to experimentally validate the in silico predictions generated by SEQSIM, particularly in regions like the PRAMEF locus where structural homology and chromatin conformation appear tightly linked. Integrating SEQSIM similarity clusters with existing experimental datasets may uncover cell-type-specific regulatory hubs informed by promoter architecture. Additionally, extending SEQSIM’s framework to include non-promoter regulatory elements, such as enhancers and insulators, could further illuminate the sequence-level rules governing 3D genome folding. On the computational side, improving the tool’s capacity to incorporate evolutionary conservation and mutational tolerance may refine its predictive power in disease and development. Ultimately, pairing SEQSIM analyses with wet-lab experiments, including CRISPR-based promoter editing and DNA-FISH, will be critical for testing the functional significance of the sequence-driven structural predictions described in this thesis.
Across all chapters, the cumulative findings challenge conventional notions of gene regulation as an isolated or solely cis-encoded process. Instead, the work supports a model in which promoter sequences participate in both linear and spatial networks of similarity, potentially reinforcing transcriptional patterns via sequence-structure coupling. The development of SEQSIM as an accessible and scalable analytic tool not only enables such investigations but also opens new avenues for the study of regulatory genomics, especially in underexplored contexts. This work builds on the foundation of structural biology, an increasingly vital field for decoding the spatial logic of genomic regulation. Previously, I contributed to a novel model of viral entry that reframed how membrane fusion events may be initiated - these and some of my previous bioinformatic work are also included in the Appendix. In parallel, this thesis expands such structural perspectives to the genomic scale. The thesis establishes that promoter sequence similarity is a widespread, informative, and structurally relevant feature of the human genome, warranting its inclusion in future models of transcriptional regulation and genome architecture. Whether protein-based or genomic, advancing our understanding of biological structure depends on the continued development of tools tailored to each scale
Improving Resuscitation of the Donation After Circulatory Death (DCD) Heart in a Female Porcine Model of Ex Situ Heart Perfusion (ESHP) Using a Combination of Intralipid, Sevoflurane and Remifentanil
No full textHeart transplantation remains the gold standard for many cardiovascular diseases, but it is limited by the pool of usable donor hearts. Donation after circulatory death (DCD) hearts are an attractive source of “extended criteria donor hearts”, but they suffer significant warm ischemia-reperfusion injury due to the circulatory death process during procurement. Currently available preservation strategies for the DCD heart during transportation are limited, and focus primarily on minimizing warm ischemia-reperfusion injury, yet they fall short of leveraging pharmacological approaches to actively enhance myocardial recovery and mitigate functional decline. As a result, there is a pressing need for innovative approaches to enhance cardioprotection during DCD heart preservation. Intralipid, sevoflurane and remifentanil are well-characterized, clinically utilized pharmacological agents known to display cardioprotective properties and to reduce ischemia-reperfusion injury. When these three drugs were given in combination in an isolated rat heart ischemia-reperfusion model, it resulted in the greatest recovery of cardiac function. Therefore, we investigated the effects of a combined cardioprotective treatment consisting of Intralipid, sevoflurane, and remifentanil applied upon reperfusion of porcine DCD hearts following procurement. The DCD hearts were perfused on an ex situ heart perfusion (ESHP) platform for 6 hours, mimicking the transport and preservation period prior to transplant into the recipient. Functional assessments of the DCD heart were performed hourly, and biochemical analyses were carried out on perfusate samples collected at various time points, as well as left ventricular tissue collected at the end of ESHP. Tissue from healthy controls not subjected to the DCD nor ESHP processes served as a healthy reference control.
DCD hearts receiving the cardioprotective treatment had improved cardiac function as measured by cardiac output, left ventricular stroke work, systolic blood pressure, inotropy and lusitropy, resulting in minimal functional decline over 6 hours of ESHP. In addition, treated DCD hearts had higher myocardial oxygen consumption over time as compared to untreated DCD hearts. Measurement of protein carbonyl content, a marker of oxidative stress, was significantly elevated in the untreated DCD heart, but was unchanged from healthy controls in the treated DCD heart. Circulating levels of cell-free mitochondrial DNA was also elevated by 8-10 fold in the untreated group during early reperfusion (30min), but was unchanged from baseline in the treated group. Further examination by transmission microscopy revealed a significantly higher proportion of damaged and compromised mitochondria in untreated DCD hearts, as well as signs of contractile dysfunction. Assessing the enzymatic activities of the electron transport chain showed a significant impairment in complexes I-III in untreated DCD hearts, but not in treated DCD hearts. As a result, the untreated DCD hearts also had an elevated NADH/NAD+ ratio as compared to treated and healthy hearts. Metabolomics screening further revealed a depletion in the latter half of the Krebs cycle intermediates in untreated DCD hearts but not treated DCD hearts. Furthermore, significant cardiotoxic triglyceride accumulation was observed in untreated DCD hearts, but not in treated DCD hearts which metabolized Intralipid as an energy substrate. Profiling of the left ventricular transcriptome by next generation sequencing also provides evidence that metabolism, particularly lipid handling, was impaired and dysfunctional in untreated DCD hearts. Notably, the DCD and ESHP process alone induced significant transcriptomic changes in DCD hearts irrespective of treatment. However, the treated DCD hearts promoted more adaptive gene programs rather than maladaptive ones in the untreated DCD hearts. Profiling of extracellular vesicle-derived miRNAs during early reperfusion also supports this notion, where the majority of miRNAs upregulated in the treated group were associated with cardioprotection and survival, but the majority of miRNAs upregulated in the untreated group was associated with cardiac injury and cell death processes. Of note, blood cells in the ESHP circuit perfusate were also a contributing source of cell-free mitochondrial DNA and extracellular vesicle-derived miRNAs in this study. Our findings shed light on the molecular changes in the DCD heart in the absence and presence of a cardioprotective treatment, allowing for better preservation strategies to be developed for the preservation and resuscitation of the DCD heart. More importantly, the cardioprotective treatment consisting of Intralipid, sevoflurane and remifentanil used in this study represents a highly promising strategy for improving the resuscitation of the DCD heart which can be rapidly translated into the clinical setting
Development and Evaluation of Crumb Rubber-Modified Binders for High-Performance Asphalt Concrete
No full textAsphalt pavements, or asphalt concrete (AC), are viscoelastic composites designed for modern transportation but often exceed their design limits, leading to distresses like rutting, moisture damage, and cracking. To address these issues, high-modulus asphalt concrete (HMAC) was developed in France in the 1960s as enrobé à module élevé (EME), using stiff asphalt binders to enhance rutting and moisture resistance. However, increased stiffness reduces stress dissipation, making HMAC prone to fatigue and thermal cracking, a primary concern in cold regions like Canada, where extreme temperature fluctuations accelerate pavement deterioration.
This thesis explores a novel approach to pavement design that addresses the limitations of hard-grade asphalt binders in HMAC, particularly in cold climates. By incorporating crumb rubber-modified binder (CRMB), a sustainable and performance-enhancing alternative, this research aims to develop high-performance asphalt concrete (HPAC). This advanced mixture retains or improves the superior dynamic stiffness of HMAC while significantly improving cracking resistance, ensuring long-term durability. Additionally, HPAC is tailored explicitly for base course applications in full-depth asphalt pavements, where tensile and compressive stresses are at their highest. Moreover, this research incorporates a comprehensive asphalt binder modification technique using a crumb rubber modifier (CRM) and its effect on AC mixtures regarding resistance to prevalent pavement distresses, such as rutting, moisture damage, and cracking.
The study is divided into five key sections. The first section introduces asphalt pavement technologies, highlighting advancements and the challenges posed by increasing traffic loads and climate change, along with the study’s objectives and methodology. The second section reviews existing literature on AC design methodologies, the rheological behaviour of conventional and modified binders, and the evolution of HMAC to HPAC. This review identifies research gaps from relevant literature and emphasizes the need for HPAC solutions suited to Canada’s extreme climatic conditions and growing traffic demands.
The third section evaluates the rheological performance of CRMB by examining the effects of crumb rubber modifier (CRM) content and blending time on asphalt binder properties, notably achieving a high Performance-Grade (PG) 82 intended for HPAC application. A 30-mesh CRM was mixed into a PG 64-22 unmodified binder at varying concentrations (3%, 6%, 9%, 12%, and 15%) and blending times (30, 60, and 90 minutes) at 180°C. Results indicated a 12% CRM content with a 60-minute blending time optimized polymerization, achieving PG 82-22, demonstrated superior high-temperature performance while maintaining adequate elasticity and low-temperature stress relaxation properties. Intermediate-temperature results suggested improved fatigue resistance, while aging analysis revealed lower stiffness gains over time, indicating enhanced resistance to oxidative and physical hardening. Consequently, the optimized CRMB suitability for HPAC application was validated with the dynamic modulus test, exceeding the requirement (≥14,000MPa @10°C, 10 Hz). However, phase separation remained a challenge, emphasizing the need for improved storage stability techniques.
The fourth section incorporates the optimized CRMB into AC mixtures to assess its suitability for HPAC applications through performance-based evaluations. Comparative analysis with unmodified AC confirmed that crumb rubber-modified asphalt concrete (CRMAC) mixtures demonstrated superior rutting resistance and moisture durability in the Hamburg wheel tracking (HWT) test and exhibited enhanced stress redistribution, effectively delaying crack initiation at low temperatures in the disc-shaped compact tension (DCT) test. The performance space diagram (PSD), developed within the balanced mix design (BMD) framework, further illustrated this improvement by positioning CRMAC mixtures in a more balanced performance zone and reinforcing their potential for base course applications in cold regions.
The final section presents the key findings of the study, including a summary, conclusions, and future research directions. The results emphasize CRMB’s potential to enhance the durability and structural integrity of HPAC mixtures, positioning them as a promising alternative for base course layers in extreme climates. While CRMAC demonstrates significant performance improvements, further research is recommended to evaluate its long-term field performance and optimize mix designs to mitigate temperature and traffic-induced distress
Application of deep learning for thyroid lobe segmentation and echocardiography fusion
No full textMedical ultrasound imaging is a portable and cost-effective real-time imaging modalitycompared with computed tomography (CT) and magnetic resonance imaging(MRI). However, ultrasound images are obtained from a handheld probe and thusare operator-dependent and susceptible to several artifacts, such as heavy specklenoise, shadowing, and blurred boundaries. This thesis presents two significant applicationsof deep learning: the automated segmentation of the thyroid lobe and thedeep network for echocardiography fusion.Clinicians manually assess the size of the thyroid gland in both transverse (TRX)and sagittal (SAG) scans as part of thyroid diagnosis. To accomplish this, the cliniciansmanually identify and delineate the thyroid gland and assess its maximumdimensions in both TRX and SAG views by scrolling through every frame of TRXand SAG scans. This task increases the workload and is time-consuming. To overcomethis manual process, an automated method for segmenting the thyroid glandfrom ultrasound scans of a tissue volume is a noteworthy contribution. A novel deeplearningnetwork is proposed to segment the thyroid lobe in ultrasound images andevaluated on the comprehensive dataset of thyroid ultrasound scans from 12 differentimaging centers, and it is benchmarked against the state-of-the-art segmentationmethods, demonstrating the competitive performance on both the transverse andsagittal thyroid images.Echocardiography scans suffer from low signal-to-noise ratio and occasional signaldropout. The scans taken from a single position of the heart have a limited fieldof view, making it challenging to visualize the heart structures. There is a need for integrating them as the images acquired from different positions lack accuratespatial alignment and require image registration for alignment correction. Theselimitations could be overcome by developing an image fusion process combining twoor more registered images from the same modality or different modalities to increasethe information and quality to advance the diagnostic value of the images. Thebenefit of using an unsupervised deep-learning approach to fuse 2D images fromhuman volunteers is demonstrated in this thesis, which outperforms other state-ofthe-art models for fusion in terms of quality metrics. The clinical significance of thismethod lies in the ability to provide accurate visualization of the heart structures,including the edges of chambers and other anatomical details which can improve thediagnosis of various heart diseases
Lori Suet Hang Lo - Abstract 65 - Innovate Conference 2025
No full textPeople living with kidney failure experience more depressive and anxiety symptoms than the general population, yet their experiences of mental health burden remain under-addressed. We looked at four chronic illness groups closely aligned with kidney failure (hypertension, diabetes mellitus [DM], multiple myeloma, and renal cell carcinoma) to identify how mental health care for adults is provided in Canada