22691 research outputs found
Sort by
Efficient Fault Injection for Exposing and Reproducing Failures in Cloud Systems
Full text linkIn today’s digital era, the reliability of cloud systems is paramount as the world increasingly depends on them for services. However, the complexity of modern cloud infrastructures, designed for high availability, fault tolerance, and scalability, makes the reliability of distributed systems a formidable challenge. Failures in these systems can lead to significant financial losses and other consequences, yet they remain difficult to detect, monitor, reproduce, diagnose, and recover from.
Faults, a common source of reliability issues in distributed systems, have attracted decades of research efforts. Many distributed system failures stem from bugs triggered by subtle faults under rare timing conditions. Furthermore, debugging failures often requires their reproduction, but reproducing fault-induced failures similarly depends on precise fault injection. This dissertation aims to improve the tools for fault-related bugs and failures, with a particular focus on bug detection and failure reproduction, to improve distributed system resilience. The core challenge is identified as how to efficiently determine the type, location, and timing of the injected fault. Firstly, a fault injection testing framework is proposed to efficiently expose partial failure bugs, by inferring and leveraging system state. Secondly, a tool is developed to efficiently reproduce fault-induced failures, by extracting and leveraging system runtime information. Lastly, this dissertation concludes by elaborating on the insights into future directions for further enhancing the efficiency of fault injection techniques to advance the state of reliability in modern cloud infrastructures
The Impact of Artificial Intelligence on the Deliberate Biological Threat Landscape
Full text linkTechnology plays a critical role in society, driving advancements that have led to lifesaving vaccines, nuclear power, and space exploration. These innovations have changed the day-to-day lives of people worldwide. However, the same technologies can be misused to pose catastrophic threats to humanity and to cause other societal and individual harms. They can facilitate the spread of disinformation, generate malicious content, and contribute to the development of conventional and unconventional weapons. Such technologies are known as being “dual-use” because they can be used for beneficial and malevolent applications. They are therefore subject to the “dual-use dilemma”, a situation in which it is difficult to prevent the misuse of a technology without limiting its benefits. The convergence of artificial intelligence and the life sciences represents the coupling of two dual-use technologies to achieve new scientific capabilities.
This dissertation explores key aspects of the governance of artificial intelligence in the life sciences with the perspective that policy is one method of risk mitigation. It analyzes the impact of artificial intelligence on mail-order DNA synthesis biosecurity measures and presents a novel framework for characterizing the impact of artificial intelligence on biological risk. Additionally, it reports on virologist perspectives from semi-structured interviews about the risks and benefits of artificial intelligence and on the development of practical policy. It concludes by discussing the current landscape around artificial intelligence and biosecurity, noting the great uncertainty of the federal government’s role in the future
INVESTIGATING ACCESS AND OPPORTUNITY FOR EMERGING MULTILINGUAL LEARNERS: A 3-STUDY DISSERTATION
Full text linkNational calls to improve educational outcomes for emerging multilingual learners (EMLs) focus on academic outcome disparities between students labeled English learners (EL) and their peers without the label. This narrative ignores the nuances of the EML trajectory. In this dissertation, I investigate how outcomes vary for EMLs over the course of their trajectory rather than simply focusing on whether they are labeled EL. In Study 1, I use multi-level Poisson regression to assess the relationship between being labeled EL and likelihood for suspension. I find that students labeled EL are more than twice as likely to be suspended compared to peers without the label, holding student and school characteristics constant. In Study 2, I used school-cluster robust inverse probability weighted regression to approximate a causal estimate of the relationship between being reclassified as English fluent by 3rd and 5th grade on English Language Arts (ELA) scores. I found a positive and significant relationship between reclassifying as English fluent by 5th grade and ELA scores. In both studies, I used longitudinal student- and school-level administrative data from 2014 to 2019 from one mid-sized urban school district in California. Finally, in Study 3, using interview data from 14 school leaders in California, I examined how knowledge about the school community informed school-level decision-making regarding EMLs. I found that principals who demonstrated a deep understanding of their school community—students, families, staff—were more likely to implement transformative policies that went above and beyond policy requirements and prioritized equity and social justice. Taken together, these studies support strengthening curricular programming over the EML course-taking trajectory so that they can receive the quality of education typically afforded to monolingual peers. Additionally, if programs cannot be immediately improved upon, the findings from this dissertation highlight the importance of supporting students to be reclassified as English fluent before they exit elementary school. Finally, the qualitative findings from this dissertation provide actionable recommendations for school leaders to implement transformative policies and programs and improve how they serve their EML population
Optimizing Implementation of Household Contact Tracing To Prevent Tuberculosis Epidemics in South Africa
No full textBackground: Tuberculosis (TB) remains a public health challenge in high-burden countries like South Africa. Household contact tracing (HHCT) offers a promising strategy to prevent TB transmission. However, implementation challenges require better understanding of underlying social and contextual determinants. Multidimensional poverty (MP), encompassing deprivations in health, education, and living standards, may influence HHCT implementation and help identify at-risk households to prioritize targeted strategies to more efficiently allocate constrained resources.
Methods: We analyzed the Kharituwe-TB cluster randomized trial of 4,417 individuals diagnosed with TB and their 6,150 household contacts in urban Soshanguve and rural Limpopo. We performed multilevel mixed-effects modeling to examine associations between MP deprivation and HHCT acceptability by individuals diagnosed with TB. We used Bayesian modeling to examine associations between acceptability and implementation reach. Using compartmental mathematical modeling, we projected the long-term impact of HHCT as change in TB incidence, burden of TB averted using disability-adjusted life years (DALY), and maximum cost per TB case that justifies cost-effectiveness.
Results: We found greater household MP deprivation to be significantly associated with lower odds of HHCT acceptability. Each 10% greater MP deprivation was associated with 23% lower odds of acceptability (95% CI: 11%, 34%) with stronger impact in urban households versus rural. Among all households, finding HHCT to be acceptable was associated with 65% lower odds of no reach (95% CI: 27%, 85%). Among households with some reach, finding HHCT to be acceptable was associated with 4.5 times greater odds of complete reach (95% CI: 2.18, 10.42). We projected enhanced and high-intensity HHCT to reduce TB incidence on average annually by 1.78% (1.47%, 2.07%) and 5.25% (4.40%, 6.06%), respectively, while also averting annual DALYs by 4.32% (3.74%, 4.93%) and 12.53% (10.95%, 14.18%). Using a willingness-to-pay of 377 (497) for enhanced and 838, $1,484) for high-intensity HHCT.
Conclusion: Multidimensional poverty is an important upstream determinant of HHCT implementation and may be used as a tool to identify at-risk households. Optimizing HHCT implementation by prioritizing targeted, context-specific strategies offers long-term epidemiologic impact to prevent TB epidemics in South Africa
ASSESSING PROTEIN FOLDING ACROSS SCALES AND DOMAINS OF LIFE
No full textProteins are the workhorses of our cells- performing many functions essential for life. To execute their designated tasks, most proteins must fold into their native conformations efficiently. Protein folding has been studied extensively for small proteins. However, extant proteomes are primarily composed of large multidomain proteins which follow more complicated folding pathways. This increased complexity leads to an increased propensity to aggregate, making them more challenging to study by traditional methods. In this thesis, I developed two methods that enable the study of larger, more complex proteins: in vitro experiments with optical tweezers and reporter assays in living mammalian cells.
Optical tweezers are a powerful single-molecule technique to study protein folding. As with most single-molecule methods, optical tweezers require small amounts of sample to collect data. However, the technique is limited to proteins that can be readily purified. In chapter 2, I describe a method I developed to overcome this constraint. By relying on a highly efficient and specific tethering system, I can directly pull-out proteins of interest from bacterial and mammalian cell lysates with minimal purification. This approach facilitates efficient screening of proteins. It is especially useful for large multidomain proteins and eukaryotic proteins which are often difficult to express and purify.
In chapter 3, I describe a method I developed using arrest peptides to detect co-translational folding and chaperone binding in cultured human cells. Arrest peptides (APs) are protein sequences that, once translated, cause the ribosome to stall. It has been demonstrated that co-translational protein folding leads to arrest release in bacteria, but the human system is much less well understood. My results suggest that AP translation stalling can be released by binding of chaperones near the exit tunnel in human cells. Chapter 4 outlines preliminary work aimed at expanding this assay to explore human proteins and enhance throughput, enabling parallel screening of multiple proteins.
Overall, the methods developed in this thesis enhance our ability to study the folding of complex proteins in vitro and in vivo. This knowledge is crucial for addressing the causes of protein misfolding linked to diseases such as Parkinson’s and Alzheimer’s
DEVELOPMENT AND EVALUATION OF PEGYLATED POLY (ASPARTIC ACID) NANOPARTICLES FOR CISPLATIN DELIVERY IN GLIOBLASTOMA TREATMENT
No full textGlioblastoma (GBM) is one of the most aggressive and common malignant brain tumors. Treating GBM is a challenge due to its highly aggressive nature and the blood-brain barrier (BBB) restriction which needs to be passed to deliver drugs to the brain. Conventional chemotherapy drugs, such as cisplatin (CDDP), have significant drawbacks such as off-target toxicity, limited ability to cross the BBB and poor brain penetration once in the brain. Nanoparticle-based drug delivery systems, combined with MRI-guided focused ultrasound (FUS), offer a promising non-invasive strategy to enhance cisplatin delivery to glioblastoma tumors. In this study, poly(aspartic acid) (PAA) was selected to encapsulate CDDP into nanoparticles due to its biodegradable nature and high drug encapsulation capacity. The surface of the nanoparticles was coated with polyethylene glycol (PEG) to increase stability in vivo and facilitate more effective penetration into brain tissue. We identified a promising brain penetrating nanoparticles (BPN) formulation that has desired physiological characteristics (diameter＜70 nm, zeta potential between -5 mV to 5 mV), high glioblastoma tumor cell cytotoxicity, and stability in biological environments. This formulation has the potential to be combined with MRI-guided focused ultrasound for efficient and non-invasive glioblastoma treatment
UBI JUS IBI REMEDIUM?: AN EXAMINATION OF INTERNATIONAL LAW REMEDIES FOR US CORPORATIONS THAT ARE VICTIMS OF INTERNATIONAL CYBERATTACKS
No full textThis thesis is a doctrinal study that analyses relevant international law, US domestic laws, and publicly available data on international cyber operations to demonstrate that applying international law to matters in cyberspace does not translate into meaningful remedies for US corporations adversely affected by international cyberattacks. It employs a theoretical framework that synthesizes the concepts of international cyberattacks and Hart's legal positivism and argues that although domestic US law could complement international law to provide remedies to US corporations that are affected by state-sponsored cyberattacks, significant limitations, and challenges militate against access to those remedies.
The study finds four main challenges that impede US corporations’ effective access to remedies for state-sponsored cyberattacks: (1) limited agency of corporations and difficulty in attributing cyberattacks to States, (2) jurisdictional and sovereignty issues, (3) political and diplomatic factors, and (4) legal interpretation and enforcement difficulties. In light of these challenges, the study proposes legal reform and policy development recommendations to (a) advocate for corporate rights and remedies in international law, (b) strengthen legal mechanisms for attribution and accountability, and (c) enhance international cooperation
“In Cobwebs, In Storms, In Chains”: The Gothic In Spanish American Short Fiction
Full text linkThis dissertation demonstrates the Gothic’s long historical presence, focusing on short-form literature during the twentieth century. While doing so, it also seeks to explore the shortcomings of genre and Gothic studies while providing a different perspective on them; a new model inspired by the work of Gilles Deleuze and Félix Guattari.
Within contemporary Gothic studies, the Gothic is a genre (a set of historically bound tropes and conventions juxtaposing the horrifying and the appealing) and a mode (the deterritorialized negative affects inspired by such tropes and conventions) but also a hermeneutic tool that focuses on how both genre and mode can abject Otherness and deal with the “unwillingness of the past to go away” (Sage & Lloyd Smith 4). This dissertation seeks to unify these three faces of the Gothic into a single entity (the Gothic Abstract Machine, the ‘diagram’ of the Gothic) and use this tool to analyze its elements within short fiction, the medium where its prominence is most discernible and pervasive across Spanish America.
My theoretical framework allows me to understand the passion and fear that the characteristically technological Argentinean modernity project inspired in twentieth-century writers (Leopoldo Lugones’ Las fuerzas extrañas and Horacio Quiroga’s Cuentos de Amor de locura y de muerte), the portrayal of Andean indigeneity as a seductive and dangerous threat to the modernizing project (La venganza del condor), the construction of national identity in permanent negotiation with Death after the Mexican revolution (Pedro Páramo and Aura), and the affect-driven voices of domestic abuse victims and animals in Tiempo destrozado. As such, this dissertation finds that the specters of miscegenation, ever-progressing technology, uncontrollable nature, colonialism, identity, domestic violence, and death herself are constitutive components of the Spanish American short-form canon.
Thus, this dissertation concludes that one cannot speak of a single Spanish American Gothic but of a multiplicity of Gothic Assemblages that run through these countries such as a Mexican Gothic, an Indigenista version of the mode, the modernista’s interpretation of it, and a Female Gothic. All these assemblages point towards the same Abstract Machine that ties them together
INTERNALIZED MULTILEVEL VIOLENCE AMONG BLACK PEOPLE LIVING IN AREAS OF CONCENTRATED DISADVANTAGE IN BALTIMORE, MD
No full textBlack people living in areas of concentrated disadvantage (ACD) in the United States experience inequitable structural, collective, and interpersonal violence that may internalize as self-directed violence (SDV). However, there is relatively little research on the internalization of multilevel violence in Black communities. This research advances the state of the art by examining a process of multilevel violence internalization within a community-based sample of Black people living in ACD in Baltimore, MD, USA.
This study takes a gender approach using multiple groups structural equation modeling to evaluate the psychometric properties of a population-based measure of SDV risk. Additionally, this study examines structural differences between females and males in a longitudinal model of internalization from perceived neighborhood disorder (PND) through exposure to interpersonal violence (ETV) to SDV. Finally, the study uses the same approach to evaluate how support network characteristics of size, density, percentage in the family, and multiplexity influence this process. Results found the 2-factor 10-item population-based measure of SDV was reliable and displayed measurement invariance and concurrent validity for females and males. The study found evidence of violence internalization such that PND at baseline predicted later exposure to witnessing interpersonal violence (WV) directly correlated with experiencing interpersonal violence (EV) that then significantly associated with SDV risk.
Finally, the study found significant gender differences in the influence of network characteristics on violence internalization. Greater support network size associated with greater SDV risk through greater ETV. However, this effect was significantly stronger among females who reported less PND. Meanwhile, network density was significantly associated with less SDV risk only among males. Support multiplexity significantly decreased SDV risk among females, however this effect did not significantly differ from males, suggesting it may offer generalizable protective effects. These results suggest that community-based interventions on SDV are practicable and may offer distinct advantages over individual-level interventions. Results also suggest that SDV research and prevention among Black people living in ACD would be more effective when considering compositional and structural differences between females and males
Studies in Multivariate Pareto Records
Full text linkThe natural and important notion of multivariate Pareto record is defined based on the dominance relation in and generalizes the classical notion of one-dimensional record in probability theory. Given a sequence of independent and identically distributed random vectors taking values in with continuous distribution function, we first investigate how the probability that the th observation sets a record varies with their common distribution. Then, for observations with independent coordinates, we study the asymptotic behavior (as ) of the Pareto frontier, defined simply in terms of the remaining records (i.e., maxima, or undominated points) at a given epoch , by deriving distributional and almost sure extreme results. These results are then extended to observations that follow a certain class of marginalized Dirichlet distributions, which exhibit negative dependence among the coordinates. For such marginalized Dirichlet observations, we also prove Berry--Esseen bounds for both the total number of records set and the number of remaining records using Stein's method. Finally, for the two models, we introduce a probabilistic approach to identify, with proofs, the asymptotic conditional distribution of the number of Pareto records broken by the th observation conditionally given that the observation sets a record