137091 research outputs found
Sort by
Improving Student Efficiency, Instructor Engagement, and Program Integrity in Online Degree-Seeking Programs
No full textThe rapid growth in interest in online classrooms has required institutions to consider developing tools to help facilitate large-scale student enrollment. However, many interventions are targeted at massive online open courses rather than online degree-seeking courses, which have different needs.
My research uses a mixed-methods approach to present students and teaching assistants’ perceptions of, and interactions with, recommendation systems in classroom discussion forums. To lay the groundwork for designing recommendation systems to incorporate educational frameworks, I explore the correlation between cognitive presence and urgent posts. Leveraging this information helps instructors to identify which posts in online discussion forums require their immediate attention. In addition, I focus on the impact of “homework-for-hire” contractors in maintaining academic integrity in the educational environment and identify how systems can be created to reduce students’ ability to participate in academic misconduct on third-party websites. This dissertation shows that intelligent system design in online degree-seeking programs promotes engagement and integrity among stakeholders.Ph.D.Human – Centered Computin
High efficiency broadband power amplifiers using SiGe HBT BiCMOS technology
No full textThe objective of this research is to investigate the design challenges of high efficiency broadband power amplifiers (PAs) using silicon germanium (SiGe) heterojunction bipolar transistor (HBT) and establish design methodologies to address these challenges. Broadband amplification with high efficiency is becoming crucial for emerging wireless broadband applications such as 5G/6G wireless communications, secured military/satellite communications, phased array radars, and high-resolution imaging. To support the system level requirement for low cost and high integration capability, SiGe HBT PAs can be a viable solution. To overcome an inherent efficiency and output power limitations of conventional distributed amplifiers for wideband applications, one part of this research provides two design examples of SiGe HBT non-uniform distributed power amplifier (NDPA) and develops novel and systematic design approaches to improve efficiency over wide bandwidth (BW). To meet the demand for highly efficient PAs with higher peak-to-average power ratio and wider channel BW for modern 5G NR-U/WLAN applications, the other part of this research establishes design guidelines for four types of series-combined transformer (SCT)-based Doherty output matching networks and provides a design example of SiGe HBT SCT-based Doherty PA for higher power back-off (PBO) efficiency enhancement.Ph.D.Electrical and Computer Engineerin
Radar Spectrum Sharing with Reactive Emitters using Reinforcement Learning
No full textThis research develops non-collaborative spectrum sharing techniques that may be used by radars against dynamic and reactive in-band emitters. Historically, each user of RF spectrum was allocated a different frequency band, allowing for interference-free operation but inefficient use of spectrum resources, because of temporally sparse use in some allocated frequency bands. The proliferation of commercial cellular services in the mid-2000s strained spectrum resources, resulting in US government-led spectrum auctions, which changed spectrum allocation to a sharing model to more efficiently use limited spectrum resources. Spectrum sharing is a well-established field of research with a bias toward collaborative techniques intended for use by communication systems. Modern radar systems have distinct requirements and capabilities to make them candidates for non-collaborative spectrum sharing. Radars can rapidly change transmission frequency, location, direction, time, polarization, and waveform code to increase the efficiency of spectrum use while simultaneously minimizing interference. Existing non-collaborative spectrum sharing techniques use reactive sense and avoid, predictive models, and online learning. Importantly, prior non-collaborative spectrum sharing research evaluated spectrum sharing performance in the presence of time-varying emitters and has been limited in its examination of the utility of online learning. This work extends prior non-collaborative spectrum sharing by assuming that time-varying emitters react to other in-band spectrum users. Discrete and continuous Markov decision process (MDP) models are developed for radar spectrum sharing with appropriate reinforcement learning algorithms to control radar waveform selection. Radar spectrum sharing performance is assessed through simulation using the Q-Learning algorithm to make spectrum access decisions in the discrete MDP model and a linear function approximation algorithm using the deterministic policy gradient theorem is used in the continuous MDP model. Simulations for both MDP models include a variety of time-varying and reactive emitters and alternative radar spectrum access strategies for comparison. Though we use MDP models with limited attributes of the environment and the radar waveform, this initial work is proof of concept that can be extended to larger models, especially the continuous MDP model. Through simulation, this research establishes that using reinforcement learning control algorithms for radar waveform selection results in an improvement of more than 25% in average rewards and over 5 dB improvement in signal-to-interference-plus-noise ratio (SINR) over the next best performing strategy across a selection of emitter types, including both time-varying and reactive behaviors.Ph.D.Electrical and Computer Engineerin
Enabling semantically richer queries over unstructured data
No full textQuerying unstructured data such as video, audio, and text is critical for domains ranging from traffic surveillance to healthcare and finance. Modern AI models (e.g., vision and language models) unlock significant potential for extracting fine-grained information from such data. However, current data systems that leverage these models primarily focus on efficiently executing semantically simple queries.
This thesis argues that enabling semantically rich queries over unstructured data requires rethinking both the query execution strategies and the query interfaces. To this end, we develop four systems that can efficiently and accurately process semantically rich queries over unstructured data.
First, we present Zeus, a video analytics system that efficiently localizes complex actions in videos using a reinforcement learning (RL)-based query executor. By using accuracy-based rewards during query planning, Zeus substantially improves efficiency while meeting user-specified accuracy targets.
Next, we propose Tracer, an adaptive query processing framework for multi-camera re-identification queries. Tracer uses a recurrent network with a probabilistic search model to optimally select camera feeds to process at each time step. Tracer significantly reduces the cost of re-identification queries on synthetically generated and real-world datasets.
We then introduce SketchQL, a visual query interface that allows users to sketch complex video moments. SketchQL maps these sketches to fine-grained video moments using a transformer model trained on synthetically generated data. SketchQL greatly enhances the usability and accuracy of fine-grained video moment retrieval.
Finally, we present Halo, a long-context question answering (QA) framework designed for domain-augmented queries. Halo incorporates domain knowledge into the QA pipeline via a Domain Hints interface, allowing users to specify structured suggestions that augment the original query. A three-stage execution pipeline applies these hints automatically and optimally, improving both the efficiency and accuracy of long-context QA.Ph.D.Computer Scienc
PopSignAI: Using Sign Language Recognition for American Sign Language Learning
No full textHearing parents of deaf children frequently struggle to acquire sufficient sign language skills to effectively teach their children. This lack of access to language during their critical language learning period may lead to reduced language skills later in life. We present PopSignAI, a smartphone bubble-shooter game that enables real-time practice through automated sign language recognition. Our 20-participant study reveals that incorporating sign recognition to support novices’ expressive signing practice in PopSignAI is more effective for American Sign Language acquisition compared to versions focused on receptive skills. We provide comprehensive documentation of PopSignAI’s development, including our recognition framework, software, dataset, and annotation system. Our system utilizes over 200,000 samples of 250 signs from 47 signers—currently the largest public isolated sign dataset—to train and evaluate a user- independent LSTM recognizer achieving 82.9% accuracy on independent test data. For gameplay purposes, the recognizer maintains 99.6% accuracy with 7ms inference time using a compact 2.5MB model. These findings and tools offer valuable insights for future sign language educational game development.UndergraduateComputer Scienc
Adaptive Causal Inference and Its Applications
No full textCausal inference is essential for understanding variable relationships and improving decision-making across domains such as policy analysis and biomedical research. In biomedical contexts, challenges like data scarcity, privacy regulations, and dataset heterogeneity complicate the development of robust causal models. This thesis introduces a novel framework for adaptive causal inference, leveraging meta-learning and transfer learning to enhance the transferability of knowledge and enable rapid adaptation to unseen data.
We address two key causal inference tasks: causal effect estimation and causal graph discovery, proposing adaptive algorithms that generalize across multi-source data and improve inference accuracy in heterogeneous settings. Applications of these methods include predictive biomedical imaging models, fair classification and policy systems, and algorithms for inferring causes of death. This work advances the development of personalized and reliable decision-making systems in healthcare and other fields.Ph.D.Machine Learnin
Conformal Prediction for Time-Series and Flow-Based Generative Models
No full textThis thesis addresses two key areas: quantifying uncertainty in point prediction models (Chapters 1 and 2) and modeling data distributions with flow-based generative models (Chapters 3 and 4). Chapter 1 extends conformal prediction to time-series data, providing prediction intervals with bounded conditional coverage gaps and demonstrating superior empirical performance. Chapter 2 builds on this by sequentially updating non-conformity quantiles to better capture time-series dependencies and introducing ellipsoidal prediction regions for multivariate time-series. On the other hand, Chapter 3 develops flow-based models using ordinary differential equations, enabling novel sample generation and likelihood estimation via a framework based on the Jordan-Kinderlehrer-Otto scheme for stage-wise training. Finally, Chapter 4 enhances the scalability of ODE-based models by introducing a local flow matching approach, improving training efficiency and distillation performance.Ph.D.Operations Researc
Digital Intervention to Support the Subjective Well-being of Women with Hormonal Imbalance or Menstrual-related Issues Through Traditional Chinese Medicine
No full textThis thesis explores how Traditional Chinese Medicine (TCM) can be integrated into a digital platform to support the subjective well-being of women experiencing hormonal imbalances or menstrual-related issues. Through cultural probes with participants and interviews with TCM experts, the study identified a need for emotionally responsive, cycle-aware, and culturally grounded self-care tools. Insights from this research informed the design of a personalized digital platform that incorporates mood and symptom tracking, TCM-based self-care guidance, and access to professional consultation. The final design reflects a hybrid approach that merges TCM’s holistic, preventative philosophies with user-centered digital design to promote emotional resilience and bodily balance. By translating traditional health wisdom into accessible, interactive features, the platform offers a new model for designing inclusive and adaptive digital health tools tailored to women's lived experiences.M.S.Industrial Desig
Assessing Micromobility as a First- and Last-Mile Solution to Improve Transit Accessibility for Low- Income Communities
No full textThis paper is an applied research paper for a master of city and regional planning; the document is structured into seven main chapters, including introduction, methodology, results, discussion, and references.This study examines how micromobility services—specifically shared bicycles and e-scooters—can enhance first- and last-mile (FM/LM) connectivity to public transit and improve access to employment opportunities in the Atlanta metropolitan region. Although micromobility has gained recognition as a flexible multimodal solution, important uncertainties remain regarding how different modes compare in terms of spatial availability, accessibility improvements, and equity outcomes across neighborhoods with varying socioeconomic conditions
Safe From the Start: Developing Pro-Social AI Training Datasets Through Data Workers' Critical Perspectives
No full textAI and ML systems are increasingly ubiquitous, with recent advances in LLMs and image generators, such as OpenAI’s ChatGPT and DALL·E, creating new urgency in future of work conversations [1, 2, 3, 4, 5]. My work explores how the massive datasets used to train these systems, collected and curated by a global workforce of data workers, come into being. Specifically, I examine what the perspective and lived experience of a data worker contributes to the data labors they perform. The perspectives of data workers who build the datasets for data-intensive systems, such as AI and ML systems, frequently goes unappreciated. Data workers have a unique on-the-ground view of the dataset and how it has been designed and developed, given that they are the executors of this work. Many of the problems we see with “biased” AI and ML systems can be traced back to issues with the dataset on which the system was trained. Consider the case of ImageNet, one of the most impactful computer vision (CV) bench-marking datasets to have been developed, facilitated by the labor of Amazon Mechanical Turk (AMT) workers (Turkers) [6]. The labels Turkers were offered to label images were based on WordNet [7], which has been in wide circulation since 2011. These labels, as demonstrated by Prabhu & Birhane, included terms that are offensive and not safe for work (NSFW), along with a host of nonconsensual pornographic terms [8]. Did the Turkers who annotated ImageNet’s entries come across these terms? Could they have alerted the ImageNet designers to problems with the use of WordNet labels before ImageNet became a critical benchmark dataset for CV systems?
Having seen the role that data workers equipped with CDL can play in positively shaping datasets, both in technical detail and sociocultural premise, I believe that building healthier, more pro-social AI and ML systems begins with intellectual partnership with data workers in dataset creation and development. My work is motivated by the role that data worker perspective can play when data workers are empowered to practice critical data literacy (CDL), as I observed during my ethnographic fieldwork with DataWorks, a combined work-training program, data services provider, and research platform [9]. CDL goes a step beyond regular data literacy, which refers to a skillset for reading and understanding data statistics and data visualizations [10]. In addition to those skills, practicing CDL requires developing a critical consciousness [11], in the tradition of Paulo Freire [12], which means being able to question how these data summaries were arrived at, what might be behind the motivation for their creation, and to whom they offer benefit. Finally, to practice CDL professionally also requires a workplace that supports this critical practice, namely in the form of encouraging workers to speak up and out about problems or concerns they have with dataset development. My overarching research question (RQ) is: what is the role of perspective in data work, and how can we incorporate the perspective of data workers as partners in dataset contextualization? The work that emerges is thus a study of why do we need better contextualization practices in data work, and what is the current state of data work annotation practices? What, then, is the relationship between critical data literacy and properly localized (or, contextualized) AI and ML systems? And finally, how we can collect and integrate more varied perspectives to relocate our AI and ML systems? Contributions: My work facilitates the development of safer, more pro-social AI and ML systems. Situated within critical data studies, the work described in this dissertation builds out approaches to the integration of worker perspective in datasets developed at archetypal production sites.Ph.D.Computer Scienc