QSpace: Queen's Scholarship & Digital Collections
QSpace is an open access repository for scholarship and research produced at Queen's University. QSpace offers faculty, students, staff, and researchers a free and secure home to preserve and present their scholarship.
Recent Submissions
Item "Nothing without us": Participation of Persons with Disabilities in the Law-making Process of Ethiopia(2024-09-12)It is common to hear the slogan, "Nothing about us without us." Recently, there has been a shift to "Nothing without us." However, globally, the role of persons with disabilities (PWDs) in the decision-making process is often devalued. The CRPD considers the participation of PWDs in public affairs, including the law-making process, a fundamental human right. In Ethiopia, there is a dearth of evidence regarding when and how PWDs are included in the law-making process that directly or indirectly concerns them. Thus, this dissertation aims to assess the participation of PWDs in the law-making process in Ethiopia. The thesis examines a case study of Directive No. 41/2015 and the Organization of Civil Societies Proclamation No. 1113/2019. I employed qualitative and exploratory research methods. According to the interview participants, there is a general improvement in the participation of PWDs in the law-making process in Ethiopia. In the first case study, Directive (2015), PWDs did participate in the enactment process; however, their participation was not up to the standards set under the CRPD. The main reasons were that enough time was not allocated for them to express their ideas in depth, they did not get a chance to review the law's final version before enactment, and most of their ideas and suggestions were not included in the law's final version, without explanation. The second case study, CSO Proclamation (2019), shows that PWDs had active participation. The research data also revealed that the participation level of PWDs in the enactment process of these case studies mainly depends on the government’s political will and commitment. Moreover, the research also found that the charity model of disability persists among legislators in Ethiopia. In conclusion, the level of PWDs’ participation was not uniform and systematic in the two case studies, which is not in line with the standards set under the CRPD. The dissertation concludes with general recommendations for the government, OPDs, and CSOs to ensure the active participation of PWDs in the law-making process in Ethiopia. Additionally, specific recommendations are provided to make Directive (2015) more beneficial for the majority of PWDs in Ethiopia.Item Surrogate-Based Optimization Using Continuous Piecewise Linear Models(2024-09-12)Nonlinear programming (NLP) and mixed-integer nonlinear programming (MINLP) are widely used in various application domains, including scheduling, transportation, and chemical processes, etc. Solving these problems to global optimality in a reasonable time is a significant challenge. However, with advancements in state-ofthe-art mixed-integer linear programming (MILP) solvers, we can effectively tackle nonlinear problems by converting them into MILPs. Continuous Piecewise Linear (CPWL) functions are popular modelling techniques capable of approximating any continuous nonlinear function. By using CPWL, we can convert NLP or MINLP problems into MILPs, which can be solved quickly and provide an initial point to warm-start the original problem. In some cases, when the solution is sufficiently accurate, this approach can eliminate the need to solve the NLP/MINLP entirely. This thesis examines the use of CPWL approximation methods to develop surrogate models suitable for solution as optimization problems. We focus on neural networks with rectifier linear unit (ReLU) activation functions, known as ReLU-NNs, a commonly used CPWL representation method. Our findings reveal that ReLU-NNs result in inefficient CPWL approximation and are less effective compared to other CPWL approximation techniques, such as Difference-of-Convex CPWL (DC-CPWL) [39]. Despite the popularity of ReLU-NN as CPWL models, the intricacies of the linear ipieces generated by these networks have yet to be fully explored. In Chapter 2, we first propose exact mathematical expressions for linear functions and linear regions of small rectifier networks. Moreover, we analyze the performance of the rectifier networks from a polyhedral perspective and introduce the three major challenges associated with these models: redundancy, degeneracy, and low efficiency. Furthermore, we explore DC-CPWL approximation and compare it to ReLU-based shallow and deep Neural Networks across four industrial case studies. Our findings demonstrate that DC-CPWL consistently yields highly efficient models without the issues of redundancy and degeneracy while ReLFU-NN representation generates less efficient models with several inefficient linear regions. In Chapter 3, the CPWL models derived from ReLU-NNs and DC-CPWL are reformulated as MILPs and applied to two optimization case studies: a benchmark function and a fuel cost minimization problem (FCMP) for a gas network. Our goal is to compare the optimization solution times when each CPWL surrogate model is integrated into the optimization framework. Additionally, to accelerate the optimization process using ReLU-NNs, we employ a pruning technique to compress the network size while maintaining model accuracy. The case study results show that pruning can significantly reduce computational time. However, using DC-CPWL as a surrogate model offers an even greater reduction in computational time compared to neural networks.Item Life Stress and Risk for Major Depression: A Latent Profile Analysis(2024-09-12)Major Depressive Disorder (MDD) affects over 280 million individuals and accounts for more years of ‘healthy’ life lost than any other medical condition. Half of all individuals with an initial onset of MDD will experience recurrences, whereas half will have only one or very few lifetime episodes. The Dual Pathway Model (DPM) proposes that major life stress may be able to distinguish risk for recurrent versus non-recurrent depression at first episode onset. Objectives were to use Latent Profile Analysis (LPA) to uncover profiles of risk in currently depressed individuals and determine whether major life stress prior to episode onset is associated with likelihood of profile membership. We hypothesized that those in a recurrent episode and those in a first-onset episode not preceded by major life stress will be more likely to belong to high-risk profiles. In contrast, those in a first-onset episode preceded by major life stress will be more likely to belong to low-risk profiles. The sample included 853 currently depressed individuals from six completed projects. Four latent profiles were identified: “Low Overall Risk”, “High Childhood Maltreatment, Low Symptom Severity”, “Moderate Overall Risk”, and “High Overall Risk”. Consistent with hypotheses, those in a recurrent episode were significantly more likely to belong to the moderate and high overall risk profiles than the low overall risk profile. Females were also significantly more likely to belong to the moderate and high overall risk profiles than the low risk profile. Contrary to hypotheses, those in a first episode without a preceding stressor were significantly more likely to belong to the low overall risk profile than the moderate and high overall risk profiles. Prospective studies are needed to make predictive claims about life stress and risk for recurrent depression. Nonetheless, the present findings have implications for the development of tailored intervention strategies with the highest chances of success based on a patient’s etiological profile.Item Surveying the Hormonome of Hazelnut Catkins During Winter Dormancy(2024-09-12)Deciduous woody perennials, such as hazelnut, undergo winter dormancy to protect sensitive tissues, such as flowers, from harsh conditions. The reproductive success of the tree is dependent on the release of dormancy under favorable conditions. To bloom, the tree must first experience a certain amount of chilling, followed by a certain amount of warmth. With global warming, many trees risk not being able to accumulate enough chilling to release dormancy. Also, when trees accustomed to warm climates are brought into cold climates, they might bloom prematurely and risk freezing damage. The latter is the case for hazelnut, recently adopted as a crop in Ontario, Canada. The present study investigates the hormonal regulation of dormancy in hazelnuts’ male flowers (catkins) by generating hormone profiles in early and late-blooming accessions throughout the dormant season. Abscisic acid (ABA), gibberellin (GA), auxin, cytokinin (CTK), their metabolites, as well as the ethylene precursor 1-aminocyclopropane-1-carboxylic acid (ACC), were measured. ABA decreased with dormancy progression, while GA increased. This correlation implies ABA is primarily responsible for dormancy maintenance in catkins and GA works antagonistically to ABA. Indeed, the ABA/GA ratio steadily decreased throughout dormancy. For the first time, CTKs have been reported to steadily increase during dormancy, implying CTKs also have an antagonistic relationship with ABA. Auxin and ethylene appear to primarily play a role in the onset of dormancy. Interestingly, early blooming accessions failed to accumulate the auxin conjugate, IAA-Asp and had higher ACC levels throughout most of dormancy. Cumulatively, the present study has generated the most comprehensive hormone profile in dormant flowers of deciduous woody perennials and has identified potential strategies for the delay of bloom in hazelnut catkins through the manipulation of hormones.Item Are Gas-rich Field Ultra-diffuse Galaxies and Dwarf Galaxies Distinct Populations?(2024-09-12)Over the past decade, significant developments in instrumentation and image searching techniques have uncovered thousands of low surface brightness objects that were previously uncatalogued. Among these are ultra-diffuse galaxies (UDGs) - objects that represent the extremes of galaxy properties. UDGs have stellar content similar to classical dwarf galaxies, but physical sizes more akin to Milky Way-type galaxies; as such, several theories for how UDGs may form differently from dwarfs have been proposed. In this thesis, we aim to constrain the different formation mechanisms in two ways. First, we compile and present the largest catalogue of optically-selected field UDGs and dwarfs with distance measurements in the Systematically Measuring Ultra-Diffuse Galaxies (SMUDGes) catalogue. We compare the UDGs and dwarfs in SMUDGes to investigate whether UDGs are a distinct population. Second, we compare the SMUDGes observations with two state-of-the-art cosmological simulations: NIHAO (Numerical Investigation of a Hundred Astrophysical Objects), which forms UDGs through bursts of star formation at early times and Romulus25, which forms UDGs from major mergers. Although formation scenarios for UDGs with these simulations are remarkably different, the present-day, global properties of the simulated galaxies are consistent with our observed sample. Furthermore, in both simulations and observations, we find no distinct difference between the UDGs and classical dwarf populations within the gas-richness vs size parameter space. The results presented in this work include the first detailed study of gas-rich UDGs in both observations and simulations.
Communities in QSpace
Select a community to browse its collections.