62588 research outputs found
Sort by
Transition Metal Molybdates as High-Performance Supercapacitor Electrodes
No full textTransition metal molybdates (TMMs) have increasingly gained significant researchinterest due to their promising electrochemical properties. However, previous studies
have given limited attention to the influence of crystal structure on the electrochemical
performance of these materials. In this study, distinct crystal structures
of NiMoO4 and CoMoO4, two of the most extensively studied TMMs, were synthesized
using a hydrothermal method to examine phase-dependent electrochemical
behavior. Our findings bring attention to the hydrate phase, especially the
hydrate-phase CoMoO4, which has traditionally been considered as a precursor
in the synthesis of annealed phases. Its exceptional performance is attributed to
its mixed charge storage mechanism and low charge transfer resistance. Additionally,
comparative analysis of CoMoO4 and NiMoO4 in identical crystal phases
highlights intrinsic differences in their charge storage mechanisms: CoMoO4 exhibits
surface-controlled behavior, whereas NiMoO4 is bulk-controlled. These insights
into the phase-specific and metal-dependent electrochemical characteristics
of TMMs underscore valuable guidelines for designing next-generation electrode
materials optimized for energy storage applications.
Furthermore, the α- and β-phases of Ni0.8Co0.2MoO4 were synthesized through
a hydrothermal reaction followed by annealing at different temperatures. This approach enables a comparative study focused on the effects of crystal structure
without the influence of different metal ions. ICP-MS analysis of the electrolyte
solutions reveals distinct Mo leaching rates in each phase, closely related to their
activation rates. The leached Mo contributes to the formation of electrochemically
active species on the electrode surface, facilitating charge storage, while the
remaining Mo within the material supports structural integrity.
These findings offer a new perspective on the role of Mo in TMM-based supercapacitors,
contrasting with the traditional view that Mo primarily enhances
electrical conductivity. SEM observations indicate unique morphology evolution
in each phase of Ni0.8Co0.2MoO4, with transformations from nanoneedle-like structures
to a crosslinked network occurring alongside Mo leaching. Furthermore, a
comparative analysis of Mo leaching and electrochemical performance in the α- and
β-phases highlights the impact of crystal structure on the electrochemical properties
of TMMs. The distinct Mo coordination environments in the α- and β-phases
affect Mo leaching rates, influencing both charge storage and long-term stability.
Specifically, the robust octahedral MoO6 coordination and an edge-sharing connection
between polyhedra in the α-phase provide structural stability and a slower
Mo leaching rate, whereas the more flexible tetrahedral MoO4 arrangement and
the corner-sharing connection in the β-phase facilitate faster activation but also
leads to greater leaching.
In addition to hydrothermal reactions, this thesis also investigates electrodeposition
of NiMo4 on electrochemically activated carbon cloth (EACC) as an alternative
approach. Initially, the study investigates how electro-oxidation can be used
to activate commercially available carbon cloth (CC), enhancing its hydrophilicity and making it suitable for electrodeposition. BET measurements confirm that this
process increases the surface area of CC, while XPS analysis indicates the introduction
of oxygen-containing functional groups (OFGs) onto the CC surface. By
adjusting the parameters during electrochemical oxidation, the quantity of OFGs
and the resulting surface area can be controlled.
Subsequent comparisons of electrodeposition on EACC reveal that both the introduced
OFGs and the increased surface area significantly facilitate the electrodeposition
rate and influence the performance of the final electrodes. Moreover, it is
demonstrated that applying a post-oxidation electro-reduction treatment removes
some OFGs without altering the surface area. Although this electro-reduction
slightly slows down the subsequent electrodeposition rate, it restores the conductivity
of the EACC, ultimately leading to an improvement in the rate capability
and overall electrochemical performance of the final electrodes.</p
Navigating Reciprocal Space: A Computer Vision-Driven Approach to Automated Microscopy at the Atomic Scale
No full text Mechanical failure is strongly influenced by atomic structures at grain boundaries.Analyzing them in reciprocal space provides valuable insights into their structure
and behavior by interpreting diffraction patterns to reveal orientation, misorientation,
and defect-related stress distributions. However, it\u27s tedious to capture an atomic-
scale image of grain boundaries. In this work, we present theoretical knowledge of
reciprocal space. We also provide information about our novel approach to automate
reciprocal space navigation on an atomic scale using computer vision methods.</p
Advancing Brain Disorder Diagnostics and Treatment through Application of Machine Learning on Neuroimage
No full text{"value":"Diagnosing brain disorders such as major depressive disorder (MDD), attention deficit/hyperactivity disorder (ADHD), cocaine used disorder (CUD), and Alzheimer’s
disease is challenging due to the significant heterogeneity in symptom presentation and the
limitations of traditional diagnostic methods, which rely on clinical observation and group
level neuroimaging. These approaches can overlook individual differences, leading to
inaccurate diagnoses and suboptimal treatment outcomes. For instance, many MDD patients
fail to respond to first-line antidepressants, necessitating prolonged trial-and-error strategies
that exacerbate patient suffering and increase healthcare costs. The lack of reliable biomarkers
for predicting treatment response further hinders the development of personalized therapies.
Functional connectivity (FC) analysis, which examines the temporal correlations
between brain regions, has emerged as a powerful tool for understanding the network
disruptions underlying brain disorders. However, the variability in FC patterns across
individuals complicates the generalization of findings. Machine learning (ML) methods offer
transformative potential by leveraging large-scale, multimodal datasets—such as fMRI and
clinical data—to uncover complex patterns and develop individualized diagnostic profiles.
These models not only enhance diagnostic accuracy but also predict treatment responses,
enabling personalized and precision medicine approaches.
This proposal aims to address critical challenges in brain disorder diagnosis and
treatment through five key objectives: 1) developing a novel ML algorithm for ADHD
diagnosis, 2) individualizing functional connections to predict antidepressant and placebo
responses in MDD, 3) identifying generalizable FC biomarkers for treatment response in
cocaine use disorder, 4) defining dementia subtypes by linking neuropsychiatric symptoms to
brain connectivity, 5) disentangling normal aging-related variations from Alzheimer’s disease
(AD) to improve the prediction of neuropathological protein expression. By advancing the
understanding of functional connectivity and leveraging ML techniques, this research seeks to pave the way for more accurate diagnoses, personalized treatments, and improved outcomes
for individuals with brain disorders. ","attr0":"abstract"
Identification of Tollmien–Schlichting Waves over a Flat Plate
No full text{"value":"This thesis explores the potential of using metamaterials as a passive flow controlmethod to delay Tollmien–Schlichting (TS) waves in a boundary layer over a flat plate.
TS waves are flow structures that initiate the transition from laminar to turbulent
flow. A recent numerical study demonstrated that phononic subsurfaces can attenuate
TS waves through elastic wave interference [8]. Based on their numerical findings,
this thesis experimentally investigates the feasibility of using phononic crystals to
attenuate the amplitude of TS waves in a closed-loop wind tunnel.
The study applies linear stability analysis based on the Orr–Sommerfeld equation.
A numerical solver is developed and validated using benchmark solutions, and spatial
instability analyses are performed to identify the frequencies at which disturbances
grow in space. Experimentally, a two-phase process is followed: (1) establishing
clean laminar flow and identifying the excitation frequency of the TS wave, and (2)
generating TS waves using a vortex street generator.
Results show a good match between theoretical predictions and experimental ob-
servations, indicating that the implemented phononic crystal structure is ready to be
placed in the test environment to further validate the effectiveness of the metamate-
rials. The findings provide evidence supporting the potential of passive metamaterial
flow control strategies for skin-friction drag reduction on aerospace vehicles.","attr0":"abstract"
MIML: Multiplex Image Machine Learning for High Precision Cell Classification via Mechanical Traits within Microfluidic Systems
No full textThis thesis presents an integrated framework for mechanical-guided, label-free cell sorting that synergizes high-speed localization with precise classification based on complementary feature modalities. The research addresses fundamental challenges in biomedical cell analysis through interconnected methodological innovations across detection, classification, and real-time processing domains.The cornerstone of this work is the development of Multiplex Image Machine Learning (MIML), a novel architecture that integrates bright-field microscopy images with cellular mechanical properties extracted during microfluidic transit. This hybrid approach achieves 98.3\% classification accuracy in distinguishing human colorectal carcinoma cells from white blood cells—representing an 8\% improvement over image-only methods. MIML demonstrates exceptional transfer learning capabilities, enabling effective classification of visually similar but mechanically distinct cells even with limited training datasets.To enable real-time applications, this research employs knowledge distillation techniques to compress a ResNet50-based teacher model into a student network with merely 0.02\% of the original parameters while maintaining robust classification performance. The subsequent FPGA implementation achieves unprecedented 14.5μs inference latency, establishing a new state-of-the-art benchmark that represents a 12-fold improvement over previous methods. Complementing this classification pipeline, a custom YOLO-based architecture optimized for high-speed microfluidic videos provides real-time cell detection and tracking. This detection framework integrates with Kalman filtering for robust trajectory analysis and extracts on-the-fly mechanical descriptors including deformation indices, velocity profiles, and transition times.By unifying these approaches, the thesis demonstrates a comprehensive system that advances label-free cell classification while maintaining high specificity, minimizing computational requirements, and operating at sub-millisecond latencies suitable for real-time applications. The integrated platform has significant implications for clinical diagnostics, cancer detection, and personalized therapies where label-free, high-throughput cell analysis is essential. This work establishes a cohesive narrative bridging high-speed cell handling with multi-modal data analysis, addressing critical challenges in microscopy-based cell classification and setting the foundation for future breakthroughs in biomedical engineering and cellular diagnostics.</p
On Using the Shapley Value for Anomaly Localization: A Statistical Investigation
No full textRecent publications have suggested using the Shapley value for anomaly localization for sensor data systems. Using a reasonable mathematical anomaly model for full control, experiments indicate that using a single fixed term in the Shapley value calculation achieves a lower complexity anomaly localization test, with the same probability of error, as a test using the Shapley value for all cases tested. A proof demonstrates these conclusions must be true for all independent observation cases. For dependent observation cases, no proof is available. </p
Tackling Emerging Data Privacy Risks in Machine Learning
No full textThe pervasive integration of machine learning (ML) and deep neural networks (DNNs) across diverse industries has driven transformative advancements in areas such as image recognition, natural language processing, and predictive analytics. However, as ML models become integral in sensitive domains like healthcare and finance, their increasing reliance on personal data raises significant concerns about data privacy and security. This dissertation addresses the critical and ubiquitous issue of data privacy risks in ML, providing a comprehensive study from both adversarial and defense perspectives.We begin by highlighting the pervasive nature of privacy vulnerabilities across all stages of ML model development and deployment, affecting models of varying complexity—from resource-constrained devices using compressed models to large-scale models in cloud environments. Recognizing that attackers can exploit these vulnerabilities to extract sensitive information, we first investigate data stealing attacks. We demonstrate that adversaries can extract high-quality training data even from highly compressed models using a quantized correlation encoding attack flow, underscoring the alarming feasibility of such privacy breaches in extreme scenarios.Relaxing these conditions, we then explore more general scenarios by focusing on Membership Inference Attacks (MIAs), a prevalent and realistic threat in various ML training contexts. Our analysis reveals that different MIAs, which are often overlooked, exhibit significantly different performances. Many existing defense mechanisms fail to protect against these variants, demonstrating poor generality and leaving models vulnerable to diverse attack strategies. To address this gap, we introduce NeuGuard, a neuron-guided defense mechanism that combines class-specific variance minimization with balanced output control. NeuGuard effectively aligns model responses between training and testing sets, providing robust protection against both neural network-based and metric-based MIAs, including label-only attacks, while maintaining high model utility and low computational overhead.Acknowledging that heuristic defenses like \textit{NeuGuard} lack theoretical guarantees and do not generalize well to broader attack types, we turn our focus to provable defense mechanisms, specifically differential privacy (DP). Traditional DP methods, such as Differentially Private Stochastic Gradient Descent (DP-SGD), are known to impose substantial utility costs and computational complexity, particularly under tight privacy budgets. To overcome these limitations, we propose Spectral-DP, an innovative approach that performs differential privacy perturbations in the spectral domain. By leveraging spectral filtering and block circulant matrices, Spectral-DP effectively controls the reconstruction error of removed components and DP-induced noise, achieving an optimal trade-off between privacy and utility. Our method significantly enhances model usability, especially under stringent privacy requirements, with transfer learning tasks demonstrating performance close to the original models.Collectively, this dissertation advances the understanding of data privacy risks in machine learning and presents effective strategies to mitigate them. By addressing vulnerabilities across different attack vectors and proposing robust defenses like \textit{NeuGuard} and \textit{Spectral-DP}, we contribute to the development of secure, reliable, and ethically responsible ML models, ensuring their safe deployment in sensitive applications and compliance with stringent data protection regulations.</p
Which Factors Are Associated with Sibling Relationships? Examining Family Functioning for Young Children with ADHD
No full textWhen one young sibling has a disorder often characterized by challenging behaviors, the sibling relationship is also negatively influenced. Young children with attention-deficit/hyperactivity disorder (ADHD) symptomology often exhibit challenges with impulsivity and behavioral self-regulation. Furthermore, parents of children with ADHD typically experience ADHD symptomology themselves in addition to high levels of stress. The primary purpose of this study was to explore whether child and parent ADHD symptomology and parental stress are significantly related to sibling relationships. Specifically, this study examined associations between parent and child functioning within sibling relationship qualities (e.g., warmth, conflict) for 38 families with at least one young child with ADHD (73.3% males, M age = 4.0; SD = .20). Multiple linear regression analyses examined associations between parent (e.g., parent ADHD symptoms and parenting stress levels) and child (e.g., child ADHD symptoms) functioning variables and sibling relationship quality. Parenting stress was the only variable significantly correlated with sibling conflict. Furthermore, the study conducted two exploratory analyses regarding whether behavioral parent training (BPT) improved sibling relationships for families raising young children with ADHD. Results of these analyses indicate a significant main effect of group, but no main effect for time or group x time interaction effect on sibling relationships for families raising children with ADHD. Clinicians supporting these families should incorporate parenting stress and conflict reduction interventions into treatment planning, which may help supplement the effects of BPT for positive family functioning.</p
Supporting Transition-Related Problem-Solving Skills for Students with Autism
No full textMathematics is an integral part of adult life in areas of employment, education, and independent living. Autistic students with concurrent intellectual disability (autism-ID) have significant needs in mathematics, and they exhibit lower rates of success in the post- secondary goals than students with other disabilities and their typically developing peers. Modified Schema-Based Instruction (MSBI) is a potential support for the mathematical needs of students with autism-ID in a transition-related environment. This dissertation used a nonconcurrent multiple probe across participants design (NCMPD) to evaluate the effectiveness of MSBI on transition-related mathematics skills with goal setting for three high school students with disabilities. The dependent variable was the number of task analysis items completed independently in the classroom setting and school-based work setting. Visual analysis, Tau-U and Design Comparable Effect Size are reported. Students showed an increase in problem solving performance in classroom and transition-related settings. </p
The Effects of a Paraphrasing Intervention on Word Problem-Solving Performance of Students with Math Difficulties and Reading Difficulties in Middle School
No full textStudents with comorbid math and reading difficulties experience challenges with word problem-solving, which requires students to utilize several cognitive and metacognitive processes, including text comprehension. This study used a single case concurrent multiple probe across participants design to examine the effects of a schema based instruction intervention with a structured paraphrasing component on the multiplicative word problem-solving performance of four students with comorbid math and reading difficulties in seventh and eighth grades. The intervention required students to paraphrase word problems, and select and use schema diagrams to solve the word problems. Visual analysis and Tau-U demonstrated that the intervention was effective for increasing the students’ performance with the SBI process, while less effective in improving student paraphrasing quality. Implications for research and practice were discussed. </p