39339 research outputs found
Sort by
Detecting Novel Concepts in Data Streams: To Infinity and Beyond
No full textAs new network attacks develop, new illnesses spread, or new topics surface on social media, many machine learning models become obsolete as they fail to recognize these novel concepts. This requires the development of artificial intelligence models capable of detecting, learning, and adapting to new concepts autonomously. In this context, novelty detection emerges as a critical method for identifying new concepts in data streams while ensuring the accurate classification of known ones. Despite growing research interest, progress in the field is hindered by challenges such as the lack of a comprehensive evaluation framework and the absence of robust algorithms that can adapt to changing data distributions with minimal human intervention. This thesis addresses these fundamental issues through four primary contributions.
First, we present a systematic literature review that establishes an updated taxonomy of the field, analyzing existing works to identify key challenges and research directions. Second, building on this review, we address inconsistencies in current evaluation methods by introducing a comprehensive framework for assessing novelty detection algorithms in multi-class data streams. We empirically demonstrate that key data stream characteristics, which are often overlooked, substantially impact algorithm performance and hinder fair comparisons across studies. Our framework formalizes these characteristics and introduces novel temporal metrics, enabling robust model comparison through a single, evolving performance score.
Third, to address the limitations of existing algorithms, we introduce CASCADE (Clustering-based Adaptive Stream Classification And Detection of Emerging classes). CASCADE integrates binary classifiers and threshold-based filters to mitigate false positives, while leveraging hierarchical clustering on custom meta-features to function effectively without dependence on ground-truth labels or extensive hyperparameter tuning. It achieves state-of-the-art performance, frequently matching or surpassing supervised approaches on diverse real-world and synthetic datasets. Finally, we demonstrate the effectiveness and advantages of novelty detection techniques in the domain of cybersecurity, using numerous benchmark datasets.
Altogether, this thesis provides a comprehensive overview of the current state of the field, establishes a rigorous methodology for performance evaluation, and introduces a high-performing algorithm that addresses critical limitations of existing techniques and matches or even surpasses supervised methods while operating without ground-truth labels or extensive hyperparameter tuning
Beta-blockers in adults with acute traumatic brain injury: a systematic review and meta-analysis of randomized controlled trials
No full textAbstract Background Traumatic brain injury (TBI) remains a leading cause of death and disability worldwide. Beyond the primary insult, excessive sympathetic activation contributes to secondary brain injury and poor outcomes. Beta-blockers may attenuate this hyperadrenergic surge and provide neuroprotective benefits, but their efficacy in improving long-term functional recovery remains uncertain. Methods We conducted a systematic review and meta-analysis of randomized controlled trials (RCTs) evaluating beta-blockers in adults with acute TBI. We searched MEDLINE, Embase, Cochrane CENTRAL, Web of Science, and ClinicalTrials.gov for RCTs comparing beta-blockers with placebo, usual care, or non-adrenergic comparators. Our primary outcome was long-term functional outcome, assessed with the Glasgow Outcome Scale (GOS) or its Extended version (GOS-E). Secondary outcomes included mortality, intensive care unit (ICU) and hospital length of stay, duration of mechanical ventilation, and adverse events. Results Seven RCTs (n = 559) met inclusion; six (n = 445) contributed data to meta-analyses. Only two trials (n = 259) reported functional outcomes. Beta-blockers did not significantly reduce the risk of unfavorable neurological outcome (RR 0.81; 95% CI 0.57–1.15; very low certainty). In contrast, beta-blocker therapy was associated with a reduction in mortality (RR 0.57; 95% CI 0.39–0.82; 6 trials, n = 440; low certainty) and a shorter duration of mechanical ventilation (–1.58 days; 95% CI –2.91 to –0.26; 2 trials, n = 85; low certainty). No effect was observed on ICU or hospital stay. Adverse event reporting was sparse, but no consistent safety concerns were identified. Conclusions In adults with acute TBI, beta-blockers did not decrease unfavorable long-term neurological outcomes but were associated with lower mortality and shorter duration of mechanical ventilation. Given the small number of trials and very low certainty of evidence, definitive conclusions cannot be drawn, and routine use cannot be recommended. A large, well-designed RCT is needed to establish the efficacy and safety of beta-blockers in this population. Systematic review registration PROSPERO CRD42024565361
Efficient Characterization and Manipulation of States of Light
No full textQuantum science and technology have seen rapid progress in recent decades, finding applications in computation, information processing, communication, and sensing. Accurate characterization and manipulation of quantum states is essential to the successful implementation of these applications. Moreover, optimal quantum state estimation is closely related to fundamental questions, such as limits on how fast information can be extracted about quantum systems. In this thesis, we present several contributions to efficiently estimating and transforming quantum states. We discuss optimal quantum state estimation, both in the regime of few available quantum systems as well as the asymptotic regime. We numerically investigate how assuming a pure-state can affect the performance of tomographic methods, and where adaptive quantum state tomography is in fact advantageous. While collective measurements have been formally considered as optimal solutions to various problems in quantum information, their implementation has been largely neglected due to experimental challenges. We present a proof-of-concept experiment realizing Massar and Popescu's optimal collective measurement for quantum state estimation, utilizing two-photon interference and polarization-dependent loss to achieve non-maximally entangled state projections. We demonstrate that collective strategies can outperform the best separable measurements. Furthermore, we introduce and experimentally implement a fast and automated optical polarization compensation scheme using liquid crystal variable retarders and rotating quarter-wave plate tomography. Desired accuracy thresholds can be achieved within a few compensation steps. Finally, we use a matrix-model approach to investigate the realization of spatial unitary transformations via sequences of thin phase gratings and free-space propagation. We numerically demonstrate that the grating parameters and propagation distances can be optimized to achieve various two-beam and three-beam unitary transformations, such as Pauli gates and the discrete Fourier transform. Through experimental and numerical approaches, this thesis offers new perspectives on methods for quantum state estimation and unitary transformation
For Cecil Taylor: A Research-Creation and Paracritical Approach to Analyzing Free Music
No full textThis creative dissertation offers a new method for analyzing Free Music (sometimes referred to as Free Jazz, Free Improvisation and Creative Music). With the musician-poet-dancer Cecil Taylor as a case study, the analysis uses a hypothetical double bass audition for Taylor’s ensemble (The Unit) to discover what traits Taylor would have desired in a bassist to more fully understand his musical aesthetic. The dissertation is comprised of six chapters. The first outlines the theoretical framework of the dissertation used to explore Taylor’s music: feminist research-creation and paracritical writing (Mackey 2018). The second chapter situates the dissertation within Cecil Taylor studies. Chapters three and four are the analyses of Taylor’s music itself, and the fifth contains the author’s poems followed by an open-ended conclusion. Chapter three begins the analysis with an explanation of how Taylor’s method of making music (Unit Structures) functions. This chapter also includes audio examples demonstrating the author’s interpretation of approaches to Taylor’s music on the double bass. Chapter four, “Letters to Cecil,” consists of hypothetical letters written to Taylor that are inspired by poet-scholar Nathaniel Mackey’s letter format from his From a Broken Bottle Traces of Perfume Still Emanate. These letters serve as the theoretical gloss that pairs with the author’s musical tracks, and together comprise the audition for Taylor’s unit. They also serve as a speculative approach to understanding both the poetics and aesthetics of his music: Percussive School of Double Bass, Bowing as Breath, Duende, Being Astral in and all Registers, and Dancing. The dissertation ends with the author’s own poems titled “For Cecil Taylor: Before and Around Midnight” that are inspired by Taylor and Mackey’s poetry. The letters, therefore, explore major themes the author has found in Taylor’s music as derived from close engagement with Taylor’s music and poetry
Re-Examining Adverse Outcomes of Chronic Kidney Disease in Ontario Adults: A Retrospective Cohort Study
No full textBackground and rationale: Chronic kidney disease (CKD) has increased in prevalence over the last twenty years in Canada, now affecting up to one in ten Canadians (3). CKD is associated with adverse cardiovascular outcomes, end-stage kidney disease (ESKD) and all-cause mortality. CKD is diagnosed through blood and urine testing and is defined as a reduced estimated glomerular filtration rate (eGFR) of 3 mg/mmol (4,5). The discrete eGFR cut-off for diagnosing CKD has been questioned due to physiologic changes in kidney function. Recent evidence suggests the CKD eGFR diagnostic threshold should be much higher in young adults (above 60 ml/min/1.7 3m2, e.g., 75 ml/min/1.73 m2), regardless of albuminuria and lower (< 60 ml/min/1.73 m2, e.g., < 45 ml/min/1.73 m2 in the absence of albuminuria) in older adults with CKD. Identifying age-specific, outcome-based threshold levels of eGFR would improve early CKD identification and treatment in the young, reduce overdiagnosis and overtreatment, and reduce inappropriate utilization of healthcare resources (6).
Objectives: To examine the association of age-specific eGFR thresholds with mortality, major adverse cardiac events (MACE), and ESKD using population-based data from Ontario, Canada.
Methods: Using linked, de-identified administrative datasets housed at ICES, we created a cohort of all adults (18-105) with at least two available eGFR measures between 90 and 730 days apart in Ontario, Canada, from January 2008 to December 2021, with a minimum follow-up of one year. We examined the associations of eGFR and age with adverse outcomes (all-cause mortality, MACE, MACE+, and ESKD) using Cox proportional hazards models and a regression discontinuity design (RDD) to identify specific risk-based eGFR thresholds in adults by age. Reclassification was used to examine event frequencies by eGFR in younger adults to determine an optimal eGFR cutoff for screening for adverse events associated with eGFR reduction.
Results: The cohort included 8,388,340 individuals (mean age 51.27, SD 17.37, 62.3% female, mean baseline eGFR 92.59 +/- 20.17) with a median follow-up time of approximately 1 year. The age distribution was as follows: 27.28% from 18-39, 50.94% from 40-65, 16.68% from 66-80, and 5.11% 80 or older. In adjusted models, the risk of all-cause mortality, MACE, MACE+ and ESKD was higher in 18- to 30-year-olds at a threshold of eGFR 80-90, with a stepwise decline in risk with age. Among those aged 80 or older, the risk of adverse events was higher only when eGFR was 50 or lower. Cox proportional hazard analysis demonstrated that younger individuals had a discrete increase in all-cause mortality and adverse cardiac events at eGFR 110 ml/min/1.73 m2. Sub-analyses for albuminuria and diabetes demonstrated similar results to the total population with modest reductions in risk; however, risk for younger individuals eGFR <90 ml/min were still present with albuminuria and diabetes. Regression discontinuity modelling did not identify a granular threshold by age for adverse outcomes; however, analyzing 3-, 5-, and 10-year risk for adverse outcomes in younger adults suggested that those with reductions in eGFR < 75 ml/min/1.73 m2 would benefit from closer monitoring.
Conclusions: This retrospective cohort study of 8.3 million Ontario individuals has provided a novel and robust assessment of optimal risk-based eGFR cut-offs for CKD diagnosis and related adverse events. However, a definitive cut-off for the eGFR threshold by age could not be identified through Cox proportional hazards and RDD. In younger adults, an eGFR below 90 ml/min/1.73m2 consistently showed a significant association with adverse events at 3, 5, and 10 years. A U-shaped pattern for adverse cardiovascular events was observed in middle-aged to older adults, where hyperfiltration was linked to negative outcomes. Examining adverse event frequency by eGFR in younger adults suggests that screening individuals with eGFR 75 ml/min/1.73 m2 or lower would be optimal. Improved risk-based determination of eGFR thresholds for adverse events associated with CKD will guide primary care providers, specialists, and policymakers and may influence clinical practice
Governing with AI : Four Actions to Build a Transformative and Resilient Public Administration
No full textThis project was made possible through the contributions of CEIMIA, the Canada CIFAR Chair in AI and Human Rights at Mila, and the University of Ottawa Research Chair in Technology and Society
Contraceptive Knowledge, Attitudes, and Practices in Algeria: A Multi-Methods Study with Women Living in Urban and Rural Areas
No full textContraception plays a crucial role in enhancing women’s health and well-being, advancing educational and career goals, improving quality of life, and contributing to a country’s social and economic development. Therefore, equitable access to contraceptive services is essential for both population health and societal development. Family planning in Algeria has long been a national priority, and the country has made significant progress contributing to increased contraceptive use over the years. However, despite these gains, recent years have seen a decline in use and a rise in unmet needs. Contraceptive behaviour, meanwhile, is shaped by various factors including individual knowledge, individual and collective attitudes, sociocultural norms, and health system-related barriers. Prevailing sociocultural norms can also render the needs of some groups invisible, leaving them underserved. Understanding lived experiences, perspectives, and contraceptive behaviours of women is essential to creating inclusive, equitable reproductive health policies and services. This thesis aims to understand the contraceptive knowledge, attitudes, and practices of women of reproductive age and to explore the contraceptive landscape in Algeria, with a focus on the facilitators and barriers to equitable access to contraceptive services. Using an action research approach, this multi-methods qualitative study is comprised of four components: a scoping review on women’s knowledge, attitudes, and practices in three Maghrib countries, in-depth interviews with women of reproductive age, interviews with key informants, and an online survey with female university students in Algeria. The findings from this research provide evidence regarding the current status of contraceptive knowledge, attitudes, and practices in Algeria, highlight the needs of underserved groups in accessing contraceptive information and service, and have the potential to inform ongoing policy and programmatic efforts
Transcriptome Analyses of Staphylococcus aureus Grown in Platelet Concentrates with Focus on Virulence
No full textPlatelet concentrates (PCs), used to treat bleeding disorders, are stored at 22 ± 2°C under agitation to maintain platelet function; however, these conditions favor growth of bacteria introduced during venipuncture. Staphylococcus aureus is a predominant PC contaminant and has been involved in septic transfusion reactions. The Ramirez’ lab has demonstrated that PC storage elicits differential regulation of bacterial virulence genes. I therefore hypothesized that the PC storage environment triggers transcriptional changes in S. aureus resulting in enhanced antibiotic resistance. Furthermore, these transcriptional changes result in enhanced virulence of S. aureus grown in PCs. These hypotheses were tested with two objectives: (1) assess differential expression of antibiotic resistance genes in S. aureus grown in PCs compared to tryptic soy broth (TSB), and (2) study the role of the NorB efflux pump on the virulence of S. aureus grown in PCs. Four transfusion relevant S. aureus strains (TRS) were grown in PCs and TSB and subjected to comparative transcriptome analyses. These studies revealed that the norB gene (encodes for the efflux pump NorB, which is implicated in quinolone resistance and negatively regulated by MgrA) was upregulated in PC-grown TRS compared to TSB cultures. Furthermore, Minimal Bactericidal Concentration (MBC) assays showed increased quinolone resistance in PC cultures of TRS versus TSB cultures. MBC and RT-qPCR assays with non-transfusion relevant strains (S. aureus RN6390, RN6390ΔnorB, and RN6390ΔmgrA) indicated that not only NorB, but also efflux pumps NorA and NorC, may be involved in enhanced quinolone resistance in PC-grown S. aureus. The impact of the PC storage environment on regulation of quinolone resistance driven by Nor efflux pumps has not been reported before. Furthermore, NorB was shown to be implicated in S. aureus virulence using a silkworm model. The use of silkworm larvae to evaluate virulence of S. aureus driven by NorB is novel and inform the research and transfusion medical communities about potential targets to minimize the risk of transfusing contaminated blood products. My research has open venues for further investigation on the role of efflux pumps in quinolone resistance and virulence in S. aureus grown in PCs, which could be used to propose mechanisms to enhance the safety of transfusion patients
Public health interventions for Non-Communicable Diseases (NCDs) in humanitarian emergencies: an overview of systematic reviews
No full textAbstract Background Non-communicable diseases (NCDs) encounter significant challenges in humanitarian emergencies due to limited resources and disrupted health systems. Systematic reviews can provide valuable insights into the effectiveness of interventions for NCDs in these settings. This overview aims to synthesize and critically appraise the current body of evidence from systematic reviews on public health interventions for NCDs in humanitarian emergencies. Methods We conducted an overview of systematic reviews following the Cochrane guidelines for Overviews of Reviews and the Preferred Reporting Items for Overviews of Reviews (PRIOR). We searched four electronic databases (PubMed, Medline, Scopus, and Global Health) for relevant systematic reviews. We extracted and synthesized data on interventions and outcomes from systematic reviews focusing on cardiovascular diseases, cancers, chronic respiratory diseases, and diabetes. We performed a narrative synthesis and used A Measurement Tool to Assess Systematic Reviews (AMSTAR 2) to evaluate the methodological quality of the included systematic reviews. Results Our search yielded 1,993 citations. Sixteen systematic reviews met our inclusion criteria and were included in the analysis. Six systematic reviews focused on multiple NCDs and combined interventions, five on cancer, five on diabetes, two on cardiovascular diseases, and there were no reviews on chronic respiratory diseases. The included systematic reviews highlighted the effectiveness of interventions such as electronic health records (EHRs) and primary healthcare for managing NCDs. Culturally tailored interventions involving community members and religious leaders were also found to be promising, particularly in improving cancer screening outcomes. Numerous barriers to implementation were also noted, and these included financial, logistical, and cultural challenges. Nonetheless, the overall quality of the included systematic reviews was low, with 81% being rated as “critically low confidence”. Conclusions This overview provides a comprehensive synthesis of the current evidence on interventions for NCDs in humanitarian emergencies. While certain interventions show promise, the quality of evidence is generally low. Future systematic reviews should prioritize implementing rigorous methods and transparent reporting. Additionally, more studies are needed to examine the cost-effectiveness of interventions, as evidence on costs and cost-effectiveness was almost non-existent. Protocol registration The protocol was registered on the Open Science Framework (OSF) and can be accessed at [ https://doi.org/10.17605/OSF.IO/HF4QN ]
Towards Generalizable Few-Shot Object Detection via Enhanced Representation Learning
No full textFew-shot object detection (FSOD), which aims to detect novel categories with minimal training examples, faces significant challenges in learning robust feature representations due to severe data scarcity. Additionally, FSOD models often struggle to distinguish objects from visually ambiguous backgrounds, restricting their generalization capability. We propose a novel FSOD framework designed to address these challenges through two key innovations. First, we introduce Wavelet‑Semantic Fusion Attention (WSFA), which enhances semantic ViT features by incorporating frequency-domain information via discrete wavelet transform, providing complementary edge and texture cues through cross-modal attention. Second, we propose the Learnable Background Prototype (LBP) that explicitly models the background patterns, significantly improving foreground-background discrimination. These contributions are then integrated into a unified single-stage transformer-based detection framework with inter-class contrastive learning. Comprehensive experiments on standard FSOD benchmarks (PASCAL VOC and MS COCO) demonstrate that our method achieves stable improvements over strong baseline methods and outperforms existing state-of-the-art approaches. This work provides a practical solution for scenarios with limited annotated data, enhancing the applicability of object detection in real-world applications