116964 research outputs found
Sort by
Methods for large-scale constrained optimization
No full textThis dissertation develops methodologies for large-scale constrained optimization problems. The dissertation is organized into four main chapters (Chapters 2, 3, 4 and 5). Each chapter is self-contained and focuses on a particular structure of nonlinear constrained optimization problem, introducing and analyzing new frameworks to address theoretical and practical challenges. In Chapters 2 and 3, we focus on decentralized optimization problems, the task of solving an optimization problem distributed across a network of agents. Each agent possesses only partial information about the global objective function and the aim is to collectively optimize the global objective function. Gradient tracking methods are one of the most popular approaches for solving decentralized optimization problems. Each iteration, these methods perform two operations (steps): (1) computing local gradients, and (2) communicating information with local neighbors in the network. The cost of these two steps can vary significantly across applications. In Chapter 2, we propose GTA (Gradient Tracking Algorithmic Framework), a unified framework of gradient tracking methods for decentralized optimization, endowed with flexibility with respect to the number of communication and computation steps performed in each iteration. We establish theoretical convergence results for GTA, which is capable of converging under any specified composition of communication and computation steps, and quantify the resulting improvements. By encompassing popular gradient tracking methods as special cases, GTA enables a direct theoretical and empirical comparison among these methods. Finally, we demonstrate the efficacy of GTA on quadratic functions and binary logistic regression classification problems. In Chapter 3, building upon the unified framework presented in Chapter 2, we introduce RGTA (Randomized Gradient Tracking Algorithmic Framework), a randomized algorithm that also balances communication and computation steps in decentralized optimization. RGTA maintains any specified composition of communication and computation steps on average and converges to the solution in expectation, resulting in a more flexible framework. We establish theoretical results on the convergence and complexity of RGTA, followed by demonstrations of its effectiveness on quadratic and binary logistic regression problems, alongside comparisons with other state-of-the-art methods. In Chapter 4, we focus on composite optimization problems, where the objective function is the sum of a smooth and a possibly nonsmooth function, with the smooth component is either a finite-sum function or an expectation of a stochastic function. Proximal gradient methods are among the most popular methods for solving composite optimization problems when the nonsmooth component has a simple structure, due to their computational efficiency and ease of implementation. When the smooth component is either a finite-sum function or an expectation of a stochastic function, it is computationally expensive or impractical to evaluate its gradient. To circumvent this issue, estimates of the gradient are utilized within proximal gradient methods. For such methods, the primary computational costs are; (1) the number of iterations and (2) the number of stochastic gradient evaluations. We propose proximal gradient methods that dynamically adjust the accuracy of the gradient estimate, progressively increasing the accuracy as the iterates approach a solution, leading to high precision solutions with minimal computational overhead. We allow for biased gradient estimates and show that the proposed methods achieve the optimal iteration complexity for first-order methods. Furthermore, when the gradient estimate is unbiased, we show that the methods also achieve the optimal complexity in terms of the number of stochastic gradient evaluations. In Chapter 5, we focus on optimization problems with a stochastic objective function and general deterministic nonlinear constraints. We propose a framework based on the Retrospective Approximation (RA) paradigm, which sequentially constructs increasingly accurate approximations of the true problem that are solved to a specified accuracy via a deterministic solver, thereby decoupling the uncertainty from the optimization. These frameworks retain the advantages of deterministic optimization methods, such as fast convergence, while achieving the optimal performance of stochastic methods without the need to redesign complex algorithmic components. For problems with general nonlinear equality constraints, we present a framework that can employ any deterministic solver and analyze its theoretical work complexity. We then present an instance of the framework that employs a deterministic Sequential Quadratic Programming (SQP) method and achieves the optimal complexity in terms of gradient evaluations and linear system solves for this class of problems. For problems with general nonlinear constraints, we present an RA-based algorithm that employs an SQP method with robust subproblems. Finally, we demonstrate the empirical performance of the proposed framework on multi-class logistic regression problems and benchmark instances from the CUTEst test set, comparing its results to established methods from the literature.Mechanical Engineerin
Principals’ stress, coping, and perceived role in addressing school stress
No full textResearch on stress during COVID-19 in schools has primarily focused on teachers and students, overlooking potential impacts on principals. This study aimed to address this gap by exploring principals’ experiences of demands and resources during the pandemic, alongside their efforts to address stress within their schools. Researchers conducted and analyzed semi-structured interviews with 14 principals using interpretive phenomenological analysis. Findings revealed principals’ diverse demands, various resources, efforts to manage school stress, perceptions of
stress in their school communities, and related health and job satisfaction outcomes. COVID-19 appeared to mainly exacerbate pre-existing demands. Implications, including potential interventions and resources to benefit principals, are discussed.Educational Psycholog
Driving patient-centered nutrition education in dietetics : identifying barriers and developing practical solutions
No full textPatients with type 2 diabetes report receiving nutrition education that lacks relevance, which hinders their ability to make the dietary behavior changes required to manage their disease. While dietitians are trained to provide patient-centered, theory-based education to support dietary behavior change, research suggests they are unable to consistently deliver this high-quality education, particularly for culturally diverse patients. A lack of appropriate education materials, insufficient time, and inadequate training hinder dietetic care, however no study has investigated how these factors influence the education provided to patients or how to address these factors in practice. Therefore, we aimed to 1) identify how dietitians currently select nutrition education, 2) evaluate the education materials dietitians use, and 3) determine how to design a clinical decision support system for patient-centered nutrition education. The first aim utilized semi-structured dietitian interviews for a qualitative thematic analysis. Resulting themes highlighted that dietitians attempt to individualize education to meet patients’ needs but lack a systematic, theory-based process. The dietitian and their healthcare environment can limit the provision of patient-centered education, often disproportionately impacting Spanish-speaking patients. Education materials could attenuate these barriers, however dietitians often lacked access to materials that adequately met patient needs. The second aim evaluated the understandability, actionability, readability, and theoretical content present in dietitians’ education materials. Although most materials were understandable, they often exceeded recommended reading levels and lacked actionable information to drive behavior change. The final aim focused on how to improve the implementation of materials to support patient-centered nutrition education. We conducted an iterative design process and dietitian user testing to prototype a clinical decision support system that could support customized material development within current practice constraints. While dietitians were eager for support, we found that customization and efficiency would ultimately determine whether such a system could be integrated into practice. Overall, these studies elucidate the complex challenge of providing patient-centered nutrition education in the current healthcare environment. However, leveraging novel technologies to improve the implementation of educational tools, namely education materials, is a promising solution to ensure all patients receive the diabetes nutrition education they need.Nutritional Science
Optimal placement of dynamic voltage restorers and static synchronous compensators for enhancing power quality
No full textThe growing demand for microprocessor-based control systems and the electrification of industrial systems have increased the need for advanced power quality (PQ) solutions to ensure reliable and stable operation. Among various disturbances, voltage sags pose serious risks to industrial processes, often resulting in equipment malfunction or production losses. This study investigates dynamic voltage restorers (DVRs) and static synchronous compensators (STATCOMs) as PQ issue mitigation devices using electromagnetic transient simulations. Following a unified modeling framework, DVRs are analyzed with a focus on voltage sag disturbance characteristics, dynamic response, and statistical evaluation of placement strategies. STATCOMs are modeled to assess voltage support capabilities, with metaheuristic optimization algorithms to determine optimal placement scenarios. Device-specific operational objectives and electrical characteristics are incorporated throughout the modeling and analysis. Simulation results demonstrate that DVRs are most effective when installed at or near sensitive loads, as they directly mitigate voltage sags at the point of utilization. For STATCOMs, single installation offered the best performance when located near the service entrance. However, in the case of multiple installations, placing the devices at or near load buses enabled more efficient voltage regulation with significantly lower reactive power requirements.Electrical and Computer Engineerin
Low rank-based image reconstruction methods for dynamic contrast enhanced multispectral optoacoustic tomography
No full textDynamic contrast-enhanced multispectral optoacoustic tomography (DCE-MSOT) is a non-invasive imaging modality that uses optical excitation along with acoustic detection for in-vivo estimation of tumor perfusion rates. The image is reconstructed by using measurements gathered at multiple tomographic views, incident wavelengths, and timesteps (frames). In practice, the number of unknown spatiotemporal pixel values in the image can reach up to 10¹⁰ making the reconstruction of such a high-dimensional image computationally and memory intensive. Therefore, developing efficient numerical solvers is crucial for the practical application of DCE-MSOT. To develop efficient accurate reconstruction methods, it is essential to exploit the underlying redundancies in the image due to correlations across the spatial, temporal, and spectral dimensions. Previous works have demonstrated that a low-rank-based representation of the sought-after spatiotemporal object can achieve accurate, memory-efficient high-resolution estimates of the dynamic object. This work leverages low-rank approaches to reconstruct multispectral dynamic images efficiently, by optimizing the computational and memory cost without compromising on the accuracy. The proposed reconstruction methods are validated and compared in virtual imaging studies employing a dynamic photoacoustic mouse numerical phantom and varying levels of measurement noise. Additionally, an in vivo tumor perfusion study in a small animal model is conducted to illustrate the proposed methods in a preclinical application. Comparisons with conventional frame-by-frame image reconstruction show that our method achieves higher accuracy and better goodness of fit in estimating the normalized tumor perfusion wash-in and wash-out rates, while requiring fewer computational resources.Computational Science, Engineering, and Mathematic
Competition as a driver of sexual dimorphism : intraspecific differences in sexual dimorphism depend on congeneric richness
No full textCompetition among species drives morphological adaptations that reduce resource overlap. However, the effect of competition on morphological differentiation between sexes within species remains poorly understood. In high competition communities, we might expect that nonsexual intraspecific morphological traits will have lower intraspecific morphological trait differences, while maintaining differences in sexual traits. In contrast, regimes with low competition may allow for higher intraspecific differences of nonsexual traits, resulting in larger divergences between the sexes. Here we use the Anolis carolinensis complex, a group of closely related lineages distributed broadly across the Caribbean, as a natural experiment to test predictions of the effect of competition on sexually dimorphic traits. We use micro-computed tomography scans to measure skull shape and size of nonsexual and sexual traits to assess if competitive regime influences the magnitude of sexual dimorphism in these two categories of traits. We find that increased competition decreases the magnitude of sexual dimorphism, with moderate to strong support across both nonsexual and sexual size traits. Shifts in size traits primarily occur in males, suggesting constraints on female adaptation to competitors. In high competition communities, skull shape was characterized by broader snouts and more robust jaw regions. Our results suggest that the magnitude of sexual dimorphism is jointly influenced by intraspecific and heterospecific competition, offering insights into morphological adaptation and species interaction.Ecology, Evolution and Behavio
Discovery and optimization of inhibitors and probes that target New Delhi metallo-[beta]-lactamase
No full textSince the discovery of penicillin in 1928, β-lactam antibiotics have been used worldwide to treat a broad spectrum of bacterial infections. However, the rise of bacterial resistance through the evolution of β-lactam hydrolyzing enzymes, known as β-lactamases, threatens their clinical efficacy. β-lactamases typically have a conserved serine in their active site that serves as a nucleophile in hydrolysis of the β-lactam substrate. Codrugs in the form of covalent inhibitors that target this nucleophilic serine are prescribed along β lactam antibiotics to counteract this form of resistance. However, an emerging class of β lactamases, known as metallo-β-lactamases (MBLs), poses a threat to this approach. MBLs instead have two zinc ions in their active site that coordinate a hydroxide molecule that serves as the nucleophile, rendering them resistant to canonical β-lactamase inhibitors. Currently, there are no FDA-approved drugs that can effectively inhibit MBLs. This is particularly alarming in the case of New Delhi Metallo-β-lactamase (NDM) which is the most widespread and clinically threatening MBL. NDM has a shallow, non-selective active site that confers resistance to multiple classes of β-lactam antibiotics including last resort carbapenems. This has placed NDM-expressing bacteria on the Center of Disease Control’s list of “urgent threats”, its highest-ranking level of concern. The emergence of NDM variants with low catalytic residue conservation makes it difficult to design reversible inhibitors that can establish strong enough interactions to effectively outcompete an administered β-lactam without driving further mutations. The use of hypothesis-driven screening and lead optimization that led to the discovery of novel irreversible covalent inhibitors of NDM is reported. Furthermore, the development and characterization of a reversible fluorescent probe that can be used to monitor the dynamic metalation, substrate turnover, and inhibitor binding of NDM in living cells is reported. The results of these experiments will aid in the characterization of NDM as well as the development of codrugs that can be used to counteract this urgent threat to global health.Pharmaceutical Science
The World Is More Uncertain Than You Think: Assessing and Combating Overconfidence Among 2,000 National Security Officials
No full textThis article analyzes more than 60,000 assessments of uncertainty made by national security officials from more than forty NATO allies and partners. The findings show that national security officials are overwhelmingly overconfident and that their judgments are especially prone to false positives. Despite having strong incentives to make accurate assessments of uncertainty, national security officials share biases that are widespread among the general public. These flaws also appear to be tractable—just two minutes of training significantly improved performance. Altogether, these findings demonstrate how national security bureaucracies can leverage insights from the decision sciences to improve cognitive performance at large scales.LBJ School of Public Affair
Do high- and low-trust news frames affect political participation intentions? : an experiment faming trust
No full textThis study tested how high-trust versus low-trust news frames of government institutions shape Generation Z’s intentions to engage in three forms of political participation—conventional (e.g., voting, volunteering), unconventional (e.g., protests, civil disobedience), and alternative (e.g., activist art, mutual-aid initiatives). In a between-subjects experiment, 301 U.S. adults aged 18–28 (N = 301) were randomly assigned to read three brief, AP-style news articles framed to convey either strong institutional trust (high-trust) or weak institutional trust (low-trust). After reading the articles, participants reported perceived institutional legitimacy, experienced moral outrage, and indicated their willingness to undertake each participation form. Although high- versus low-trust framing did not produce significant direct differences in participation intentions, path analyses using parallel mediation revealed distinct cognitive and affective mechanisms: (1) low-trust frames reduced perceived legitimacy, which in turn lowered conventional participation intentions; (2) low-trust frames increased moral outrage, which in turn elevated unconventional participation intentions; and (3) for alternative participation, perceived legitimacy and moral outrage jointly operated as parallel mediators—low trust decreased legitimacy and increased outrage, each independently boosting the likelihood of engaging in alternative actions. These findings demonstrate that high-trust and low-trust framing exert their effects not through simple main effects but via simultaneous, parallel cognitive (legitimacy) and emotional (outrage) processes, and they validate “alternative participation” as a third, conceptually distinct mode of civic engagement among young adults with distinct emotive and cognitive pathways.Journalism and Medi
Heterogeneous bubble nucleation on mafic crystals in rhyolite magma
No full textThe nucleation of H₂O bubbles within magma plays a significant role in how volcanic eruptions occur and how magmas evolve through time. Recent studies propose that bubbles that nucleate heterogeneously on crystals within a magma may account for most if not all bubbles that form during an eruption. To better constrain the conditions under which heterogeneous bubble nucleation may occur on common minerals found in magmas, we ran decompression experiments on high silica rhyolite magma with crystals of augite and hornblende to observe the kinetics of bubble nucleation We discovered that both augite and hornblende are nucleation sites for bubbles at supersaturations as low as ~10 MPa and 5 MPa, respectively. Nucleation at such low supersaturations indicates these mafic phases enable heterogeneous bubble nucleation similar to magnetite. The number of bubbles nucleating on crystals increases with supersaturation. Bubbles also nucleate closer together at higher supersaturations. Crystals with more surface area have more bubbles nucleate on them compared to smaller ones at the same supersaturation, however; the number of bubbles that nucleate on a given area of crystal surface remains relatively constant for any size crystal at the same supersaturation.Earth and Planetary Science