26419 research outputs found
Sort by
Machine Learning for Anomaly Detection in Neural Network Security and SRF Cavities
No full textThis dissertation explores the development and deployment of machine learning approaches to address critical challenges in anomaly detection across two distinct domains: neural network security in federated learning settings and cavity behavior analysis in particle accelerator operations at Jefferson Lab in Newport News, Virginia. Anomaly detection identifies deviations from expected patterns, safeguarding systems in cybersecurity, industry, and research against malicious activities and failures. This dissertation demonstrates how our machine learning approaches enhance detection accuracy and efficiency in both neural network security and industrial applications.
First, we investigate vulnerabilities in deep neural networks deployed in federated learning. Although federated learning preserves user privacy by training models locally, it remains vulnerable to backdoor attacks, in which malicious participants embed hidden triggers that induce targeted misbehavior. We propose a self-supervised contrastive learning framework to detect and mitigate such backdoor attacks. In our experiments, this method achieves higher detection accuracy and lower false positive rates than existing defenses, while operating without access to local model updates or original training data and thus preserving the privacy guarantees of the federated setting. Second, we address the operational reliability of superconducting radio-frequency (SRF) cavities at the Continuous Electron Beam Accelerator Facility (CEBAF). Our research leverages an unsupervised learning approach, combined with Principal Component Analysis (PCA) and k-means clustering, to identify anomalous behaviors in SRF cavities. Our method detects subtle anomalous behavior by analyzing SRF signal data. This knowledge allows for the early detection and resolution of potential faults, significantly improving the efficiency and reliability of operations. Third, we extend these insights to time-series anomaly detection more broadly. We design a contrastive-learning based model tailored to increasingly dynamic environments and academic research. This model improves detection accuracy in settings that require real-time monitoring and predictive maintenance.
Our research underscores the broader applicability and impact of advanced machine learning techniques in anomaly detection. By extracting meaningful patterns from complex data, machine learning can significantly enhance security in distributed neural networks and improve the efficiency of particle accelerator operations. This dissertation serves as a stepping stone for future investigations into the vast possibilities of anomaly detection, inspiring further exploration and development of machine learning techniques in this field
Self-Sustained Tumbling of Bluff Bodies at Low Speeds
No full textMany objects such as low length-to-diameter ratio cylinders and some planetary entry capsules are known to experience spontaneous and/or self-sustained tumbling rotations about a transverse axis. This study quantifies the tumbling frequency and the associated lateral aerodynamic (Magnus) force as well as increased drag for a range of geometries in subsonic flow. Experiments were conducted using a unique wind tunnel magnetic suspension system, hence are free of support interference. All non-spherical geometries exhibited sustained tumbling behavior with the tumbling frequency proportional to flow velocity. It was found that certain geometries exhibit multiple stable lock-in frequencies
Understanding Executive Order 13583’s Impact to Federal Workforce Diversity & Employee Perception of Federal Policies and Programs
No full textCreating a more diverse, equitable, and inclusive (DE&I) workforce produces beneficial organizational outcomes, such as decreased turnover, improved creativity and innovation, and improved organizational performance (Diaz-Garcia et al., 2013; Gomez & Bernet, 2019; McLeod et al., 1996; Singh & Selvarjan, 2013). While the private sector has placed much emphasis on improving DE&I in their hiring practices, whether the federal government has achieved comparable outcomes remains unclear. Guided by the punctuated equilibrium theory (PET), this dissertation examines the impact of Executive Order 13583 on both change in workforce composition, and employee sentiment towards diversity policies and programs.
Using a multi-level, multi-method, quantitative analysis, this study integrates macro, meso, and micro-level analyses. At the macro-level, the interrupted time-series found no immediate impact of EO 13583 on workforce composition. At the meso-level, a panel data analysis revealed a 2.1% increase in workforce diversity across agencies. At the micro-level, using FEVS survey data, an ordered logistic regression demonstrated that federal workers had substantially more positive sentiment towards diversity policies and programs after EO 13583, with job satisfaction and management level as the largest predictors. Ultimately these findings demonstrate EO 13583 enabled only marginal changes to workforce composition, but did have robust impacts on employee sentiment towards DE&I initiatives
The Purpose and Value of a Summer Camp for Visually Impaired Young People
Get PDFEmpirical research documents the benefits of summer camps for young people, including disability-specific or medical-speciality residential camps. Using an ethnographic approach which utilized observation and individual and group discussions with the visually impaired young people who attended a summer camp, their parents, and school teachers who staffed the summer camp, we build on the extant research here by exploring, for the first time, the purpose and value of a summer camp for visually impaired young people. The qualitative data generated from our research were subjected to thematic analysis. We discuss the summer camp in relation to the following themes: (1) The summer camp facilitates peer interactions and relationship development; (2) the mixing of age groups facilitates the development of life skills; and (3) the summer camp supports the recruitment of visually impaired young people to Fieldway School [pseudonym]
A Mixed-method Analysis of the News Media Framing of Gender Non-conforming Victims of Homicide in the U.S. from 2012-2022.
Get PDFRecent analyses of transgender homicide victims find that the news media often uses improper terminology, delegitimizes, and victim blames them. These analyses, while insightful, are limited as they have largely analyzed cases involving trans women and trans feminine individuals. The present study employs a mixed method approach to analyze news media articles (N=88) published in U.S. online news media outlets about 17 gender non-conforming victims killed between 2012 and 2022. We found that most articles did not delegitimize or victim-blame. However, we find 1) victim blaming occurred when reporting on cases of officer-involved shootings, 2) certain victims receive more coverage and support, 3) confusion about terminology when discussing the gender of victims, and 4) episodic framing of transphobic violence. Implications and potential areas of research and practice are provided to address, challenge, increase, and improve news media reporting of these victims
Feel Bad to Discard a Fashion Product: How AI Designers Influence Individuals\u27 Sustainable Consumption
Get PDFThis study explores how AI technology in fashion design influences consumers\u27 sustainable consumption behaviors, focusing on emotional attachment to products. By comparing AI-generated and human-designed fashion items, the study examines how designer type impacts negative emotions about discarding products, mediated by emotional attachment. Results from two experimental studies reveal that designer type significantly affects negative emotions toward discarding human-designed items, but emotional attachment was not influenced by designer type in the first study. This lack of difference may be due to personal characteristics that moderate the effect. The second study found that individuals who perceive AI as human-like form stronger emotional attachments to AI-generated products, which impacts their sustainable consumption behavior. The findings suggest that fostering emotional connections with AI-designed products could encourage more sustainable consumer behaviors in fashion
GAN-based Event-level Inverse Mapper (GEIM) - An Application on Quantum Chromodynamics Global Analysis
No full textThe inverse problem, aiming at determining the unknown cause given an observed effect, is a fundamental challenge in scientific investigations. In the field of high-energy physics, understanding the complexities of quantum chromodynamics (QCD) relies on analyzing multi-dimensional quantum correlation functions (QCFs), which are derived from experimentally observed events. While the mapping from parameters to observable events in QCFs is a well-posed problem with unique solutions, similar to a general inverse problem of deriving parameters from observables, the inverse problem of inferring parameters from observed events, poses unique challenges due to its ill-posedness. This paper introduces a machine learning-based framework based on generative adversarial networks (GANs), the so-called GAN-based Event-level Inverse Mapper (GEIM), which is designed to address the inverse problem of femtoscale imaging in QCD. GEIM consists of two GANs: the conditional GAN-based \textit {surrogate event generator}, which replaces the physics-based QCF model to generate synthetic events, and the \textit {outer-GAN}, which performs the backward mapping to derive the parameter distributions. Through a proxy 1D QCF analysis, we demonstrate the efficacy of GEIM in accurately learning the mapping between observable events and QCF parameter spaces, deriving QCF parameters from event-level analysis, and eventually reconstructing QCFs
Rapid Prediction of Coastal Flooding with Deep Neural Networks
Get PDFWith the increasing impact of climate change and relative sea level rise, low-lying coastal communities face growing risks from extreme storm tides and recurrent nuisance flooding. Thus, timely and reliable predictions of coastal water levels are critical to resilience in vulnerable coastal areas. Over the past decade, enormous efforts have been made to utilize machine learning (ML) based data-driven models for the emulation and prediction of storm tides. However, flood advisory systems still rely on running computationally demanding real-time hydrodynamic models. because developing highly reliable ML-based models suitable for real-time forecasting and capable of capturing any surge levels is challenging. While ML-based models are very fast, challenges lie in ensuring reliability and ability to capture any surge level. In this research, we develop a deep neural network for spatiotemporal prediction of water levels in coastal areas. Our model relies on data from numerical weather prediction models as the atmospheric input and astronomical tide levels, while its outputs are time series of predicted water levels at several tide gauge locations. We utilized a CNN-LSTM setting as the architecture of the model. The CNN part extracts the features from an hourly sequence of gridded wind fields and fuses its output to several independent LSTM units. The LSTM units concatenate the atmospheric features with respective astronomical tide levels and produce water level time series. As an effort to fill a knowledge gap, we prioritized physical relation in the model by maintaining a high analogy to hydrodynamic modeling, either in the network architecture or in the selection of predictors and predictands. The results show that, with an average RMSE of 10.6cm, this setting yields a strong performance in predicting storm tides from minor to major levels
The Strategic and Governance Implications of Solar Radiation Modification: Perspectives from Delegates of International Climate Negotiations
Get PDFThe lack of progress in addressing climate change has led to increased interest in solar radiation modification (SRM)—a collection of large-scale interventions that cool the planet by managing the amount of solar radiation that reaches the earth. SRM complicates climate change governance because, in addition to advancing collective action to reduce greenhouse gas emissions, governance needs to restrain unilateral SRM action while balancing diverging actor interests, ethical risks and scientific uncertainty. We survey international climate policy experts for their assessments of the potential for effective global governance of SRM and the likelihood of possible international responses to unilateral SRM scenarios. Experts are pessimistic about the global community achieving effective SRM governance, and they believe unilateral SRM action will trigger international responses and conflicts. Experts believe softer responses are most likely (e.g. diplomatic sanctions) but the potential for stronger responses, including military action, are non-trivial. Relative to the Global North, experts from the Global South are relatively more supportive of SRM, including the development of SRM, the inclusion of SRM in international negotiations, and the deployment of SRM in a climate emergency
Accept or Reject? Moderators That Influence the Decision-Making Process of Interruptions
No full textInterruptions are a common occurrence in the workplace but when they happen during high-stakes critical tasks they can have serious consequences. In the field of healthcare, it has been shown that interruptions can lead to serious adverse events or medication errors (Kukielka et al., 2019; Makary & Daniel, 2016). An interruption consists of suspending progress on one task to address a secondary task where there was an intent was to complete the initial task. The interruption process begins with a signal (e.g., an alarm or co-worker initiating conversation). After the signal is processed information included with the signal is interpreted (Latorella, 1999; Sarter, 2013; Woods, 1995). Finally, a decision is made to either reject the interruption and continue working on the primary task or accept the interruption and address the requirements of the new task/activity. Research on interruption management strategies address the potentially harmful outcomes of interruptions but often fail to account for the initial decision to accept or reject the interrupting task, and when they do, they rarely address moderators that influence the decision.
The current study used a healthcare paradigm to examine the interruption decision-making process. The decision to accept or reject an interruption was investigated using three moderators: priority, cost of the interruption, and method of the interruption. Participants’ primary task was to monitor two patient EKG displays while simultaneously entering medication information into a patient chart. While working on the primary tasks, participants were interrupted four times throughout the experiment. Participants were assigned to groups where their interruptions were manipulated by one of three moderators. Priority of the interrupting task could be high or low. Cost was manipulated by the need to perform a task located inside the laboratory or located outside the lab far away. The method of interruption was either trigged by an alarm or having the experimenter make the request in person (face to face). After the experiment, participants filled out the NASA TLX (Hart & Staveland, 1988) as a subjective workload measure.
Sixty undergraduate psychology students with no formal healthcare experience were recruited. The decision to accept or reject an interruption was investigated using the Cochran’s Q test. There was a significant difference for decision-making in the priority condition where low priority tasks were rejected more than the high priority tasks. In addition, high-cost interruptions were rejected more often than the low-cost interruptions and face-to-face interruptions were accepted more often than the alarm interruptions. Multiple one-way ANOVAs were run to investigate differences in subjective workload among the interruption conditions. Participants in the cost condition thought they performed better those participants in the priority condition and those in the priority condition felt more frustrated than those in the cost condition. Understanding what influences the decision-making aspect of interruptions is the first step to design a system where life threatening emergencies are communicated by means in which they will be likely more accepted. Separating these interruptions can help reduce the attentional resources needed deciphering the interrupting signal and instead redirect attention to safety-critical tasks