48440 research outputs found
Sort by
Text-Centric Representation Learning: Integrating Multimodal Knowledge to Overcome Labeled Data Scarcity
No full textAdvancements in textual representation learning often face the challenge of requiring extensive labeled data, which is resource-intensive and impractical in dynamic fields like fake news detection and social media analytics. This dissertation introduces a paradigm shift from traditional "text-only" approaches to a "text-centric" methodology, integrating auxiliary knowledge from diverse sources and modalities to enhance text representation without relying heavily on labeled data. The first contribution emphasizes incorporating human prior knowledge and domain expertise. A novel framework employs multi-source domain adaptation and weak supervision to detect fake news early, transferring knowledge from well-labeled source domains to target domains with limited labeled data. Additionally, meta-learning and contrastive learning techniques are utilized to reduce noise in augmented data, improving text classification by reweighting and refining feature representations. Furthermore, we explore the use of large language models to generate high-quality augmented data, enhancing the diversity and robustness of training datasets. The second contribution explores the fusion of numerical features with textual data in causal analyses of fake news dissemination on social media. A causal inference model combining textual and numerical covariates identifies key lexicons and posts that drive misinformation spread, informing more effective intervention strategies. To achieve this, we introduce a structural causal model (SCM) [86] that disentangles confounding effects and provides deeper insights into the propagation mechanisms of misinformation. The final contribution presents two innovative approaches that integrate graph structure infor- mation with natural language understanding. First, we propose GRENADE [82], a graph-centric self-supervised learning model that synergistically combines pre-trained language models and graph neural networks to enhance node representations. GRENADE leverages graph-centric contrastive learning and knowledge alignment techniques to capture both textual semantics and structural dependencies, enabling effective representation learning without requiring labeled data. Second, we introduce AskGNN, a framework that augments large language models with graph in-context learning, incorporating graph-based reasoning to enhance the integration of textual and relational insights. This method effectively captures high-order dependencies while maintaining scalability and adaptability across various applications. Through these contributions, this dissertation advances machine learning research by proposing innovative techniques to overcome data limitations, enhance misinformation analysis, and improve graph-based text representation learning. The findings offer significant implications for applications in fake news detection, social media analysis, and natural language processing, contributing to the broader goal of developing more reliable and interpretable AI systems
Providing Chatbot Assistance to Ukrainian Refugee Women in Romania
No full textAfter the Russian invasion of Ukraine, millions of Ukrainians have had to seek refuge in Romania. The goal of this project was to assist Ukrainian women refugees in finding essential resources and information by evaluating the potential of an AI-powered chatbot on Code for Romania’s platform, Women Center. Our collaborator’s long-term goal is for this chatbot to be a tool that can be applied to various humanitarian contexts and solutions. We accomplished this goal by identifying refugee needs, interviewing experts, prototyping chatbots, and gathering data and feedback. Code for Romania is continuing this project with one of the developed prototypes guided by our design principles, prototypes, and website recommendations
Optimal Defect Layer Position in Electromagnetic Energy Absorbers
No full textBeamed energy applications use susceptors to absorb applied electromagnetic radiation and convert it to heat, which can be used to produce useful work. In this project, we consider a susceptor made of a composite material, composed of alternating high and low-permittivity layers. We explore the effects of wave-geometry interactions in this material on energy transmission to the heat exchanger. It is demonstrated that these interactions can be controlled, and reflection at the heat exchanger surface can be minimized by varying only relative layer width and free-space wavenumber. The proposed approach works for both idealized and practical material parameters
Development of a Digital Hub for Greek Roma Educational Inclusion
No full textThe project goal was to create an online hub to support the educational integration of Greek Roma students at risk of school dropout. Interviews and surveys with educators in Roma communities revealed that stereotypes, socioeconomic barriers, and expectations of early adulthood alienate Roma students from academia. Educators noted that additional support, training, and resources were necessary to create inclusive classrooms. In response, the online hub was developed featuring educator advice, infographics, and inclusive teaching resources. Usability and maintainability testing informed updates to the platform and revisions to the maintenance user guide. Recommendations include expanding the hub’s content, developing classroom materials, and building a teacher network
Investigating Potential Markers of Ferroptosis from Artemisia annua in Breast Cancer Cells
No full textBioactive compounds from the Artemisia annua plant, including artemisinin and its derivatives, may induce cell death in cancer cells via ferroptosis, an iron-dependent cell death mechanism. Using immunoblotting, we identified markers of ferroptosis in T-47D and MCF-7 breast cancer cell lines in response to different doses of A. annua tea, including increased Transferrin Receptor 1 (TfR1) expression and decreased ferritin heavy chain (FTH1) expression. MTS assays used to determine live cell numbers produced inconclusive results, and further research and experiments are necessary to determine the impact of A. annua tea on cancer cell viability
Memory Rescue in C. elegans with Alzheimer’s Disease Through Restoration of the Gut Microbiome
No full textAlzheimer’s disease is a progressive incurable neurodegenerative disorder characterized by the accumulation of beta-amyloid plaques and tau tangles leading to neuron loss and cognitive decline. Alzheimer’s severity is associated with memory loss which can be measured through associative learning assessments. Caenorhabditis elegans (C. elegans) are a reliable model organism for studying neurodegeneration. Recent studies have highlighted the gut-brain axis and the connection between neurological health and the gut microbiome, leading to the development of CeMbio, a mix of bacteria meant to simulate the natural microbiome of C. elegans. CeMbio has been shown to restore some of the motility lost in C. elegans expressing beta-amyloid plaques. This study tests if CeMbio is able to restore associative learning in C. elegans with Alzheimer’s disease through restoration of their gut microbiome
Impact of Architectural Design on Human Behavior and Emotions
No full textThis project continued previous research in the Personal Emotional Augmented Controlled Environment or PEACE room by running the experiment and creating a Python tool trained on previously collected data to analyze future data. Other aspects of the project include designing a new experiment to take place in a multipurpose room in the Center for Wellbeing on campus that focuses on lights and sound as well as designing a structural system for a new shape and size changing room technology
Vitamin Transporter Engineering for Enhanced Growth and Protein Production in S. cerevisiae
No full textThis study explores the effects of expressing mammalian solute carrier (SLC) transporters SLC 5A6 and SLC 52A3 in Saccharomyces cerevisiae to enhance growth and protein production. These transporters import Vitamins B7 and B2 respectively. Engineered S. cerevisiae strains with Green Fluorescent Protein and varying promoter strengths were tested in synthetic media with different vitamin concentrations. SLC 5A6 expression significantly boosted growth at higher Vitamin B7 levels without reducing per-cell protein output. SLC 52A3 also improved growth, though yeast’s natural Vitamin B2 synthesis limited the effect of external supplementation. These results show that Solute Carrier transporter engineering can improve nutrient uptake, growth rate, and bioproduction in yeast cell factories
Song Slicer: An Application for Evaluating and Implementing Music Structure Algorithms
No full textSong segmentation algorithms break songs into distinct parts (chorus, verse, etc.). This is useful for many types of users to better understand structure. Our project aims to address the difficulty of using current segmentation technologies by developing a musician friendly application. To choose which algorithms to include, we evaluated segmentation algorithms. Users can run different algorithms and edit segments within the application refining the segmentation process to suit their needs
Analysis, design, and application of improved high-speed mechanisms for middle-ear characterization under high-acoustical loads
No full textExposure to high acoustic pressures is a major cause of ear damage, specifically Tympanic Membrane (TM) injuries. Ongoing research must effectively load and capture TM response under high acoustic pressures, which is challenging for human studies. A previously developed ultra-high-speed shock tube generates controlled acoustic shock waves to load TM samples, with 3D High-Speed Digital Image Correlation (3D-DIC) and Schlieren imaging for response and far-field metrology. This project aimed to enhance TM investigations by increasing initial shock pressure, reducing secondary wave generation, and minimizing sample preparation effects for 3D-DIC. Using the Rankine-Hugoniot relations, we performed parametric numerical analyses to develop and experimentally validate a refined shock tube design capable of reducing secondary vortex overpressure by 87%. Additionally, a methodology was developed to study DIC sample preparation by testing samples with various speckle patterns and compounds while minimizing measurement uncertainties. These improvements are expected to contribute to further understanding TM mechanics under high acoustic loads, aiding the development of hearing protection and enhancing middle-ear surgical methods