137091 research outputs found
Sort by
Brands in Crises: Consumer and Investor Reponses
No full textCrises happen more frequently than ever. Chapter 1 explores how consumers respond to brand crises. Prior research discusses strategies that a brand may use to respond to a crisis arising from a transgression associated with the brand. When a brand uses these strategies (e.g., compensating affected parties, offering apologies), it implicitly or explicitly acknowledges its responsibility for a transgression. In many instances, however, a brand may want to deny responsibility for a transgression. For instance, if a brand’s supplier rather than the brand engaged in unethical labor practices, the brand could claim that it is not responsible for the transgression. Across four experimental studies, I find that a brand’s denial of responsibility leads to lower brand evaluations. Consumers view a brand’s denial of responsibility as morally unacceptable, even when the brand’s supplier rather than brand engaged in a transgression. The negative effect of denials on brand evaluations are moderated by consumers’ political orientations and a brand’s prior social behavior. Specifically, the negative effect is stronger among liberal consumers as compared to conservative consumers for values-related crises (but not for performance-related crises). In addition, a brand’s prior social behaviors raise consumer expectations of moral behavior from the brand more than they provide the brand a moral license for future transgressions. In Chapter 2, I investigate how investors react to firm’s decisions related to product recalls. Prior research demonstrates that investors react to timing of recalls and investigation durations (Astvansh and Eshghi 2023; Eilert et al. 2017). However, I study resolution dormancy, which I define as the time elapsed between when a recall is officially issued (either voluntarily or involuntarily) and when consumers are notified. During the dormancy period, manufacturers can prepare for the recall and manage their costs. I explore how investors respond to dormancy periods depending on the type of CSR. Utilizing data from the National Highway Traffic and Safety Administration and CSRHub, this research examines automotive recalls between 2009 and 2019. I find that investors respond negatively to dormancy periods when the manufacturer has high philanthropic CSR, as the manufacturer may not signal resource efficiency. Findings of this research provide a different perspective on understanding investor reactions to automobile product recalls.Ph.D.Managemen
Towards Addressing Some Fundamental Challenges with Brain-Computer Interfaces: A Systems Approach
No full textThis research tackles key challenges in Brain-Computer Interface (BCI) systems. Non-invasive EEG-based BCIs are the most widely used due to their safety and accessibility, holding nearly 60% of the market in 2024. However, they face persistent barriers: low signal-to-noise ratios caused by skull attenuation and unrelated brain activity, and strong non-stationarity both within and across users, requiring repeated calibration. Further obstacles include limited publicly available datasets for training modern models, constrained usability due to complex acquisition protocols and a lack of broadly applicable real-world use cases. Together, these issues restrict the EEG-based BCIs from achieving the ubiquity of modalities such as speech or text.
To overcome these limitations, we adopt a systems-level approach, viewing BCI challenges as interconnected subproblems within the broader ecosystem of signal acquisition, decoding, generalization, and usability. This perspective emphasizes exploiting resources rather than working in isolation, like data from other users, predictions from existing models, or mutual information from complementary neural signals. To promote generalization and to overcome data scarcity, we employ data-efficient techniques such as few-shot learning, active learning, meta-learning, optimal transport, etc. We also extend this principle to multi-human learning, where predictions from diverse users are aggregated to stabilize error-related potential (ErrP) classification. Beyond human resources, we introduce Response Coupling, an approach that integrates two neural signals, leveraging their mutual information to enhance accuracy, suppress noise, improve synchronization, and enable unsupervised signal quality estimation. Finally, we explore resource-aware zero-shot learning across signals with preliminary results and talk about current challenges and future research opportunities. Collectively, these contributions highlight how systematically identifying and exploiting resources leads to more robust, generalizable, and practical BCIs, moving beyond incremental improvements toward versatile real-world systems
Reimagining Play: A Creative Approach to Reducing Screen Time
No full textIn today’s world, technology is ubiquitous, providing children with constant access to interactive screens. However, excessive screen time posed significant challenges. This study aimed to investigate whether integrating a physical product—a fidget cube—into digital gameplay could effectively engage children in activities that are not exclusively digital. The intervention was designed to promote physical activity and interaction with tangible objects, contrasting with the purely screen-based nature of traditional video games. Unlike traditional video games centered around screen use, this product promoted active engagement through a pre-existing fidget toy, incorporating insights from existing literature on screen habits and parental perspectives obtained through interviews. A co-design workshop refined the design directions, resulting in Fidgital-Play—a game that integrates spatial reasoning and physical interaction with digital gameplay.
User testing involved children aged 8-10 years in 20-minute gameplay sessions, followed by a five-question survey. Results showed high engagement, with four out of five users rating the game 5/5 for fun and three eager to play again. Feedback highlighted the game’s balanced challenge and the potential to reduce screen dependency by shifting focus between digital and physical components. These findings suggest that Fidgital-Play effectively balances screen time and offers a promising approach to engaging, educational play.M.S.Industrial Desig
Supporting Effective and Trustworthy Data Communication through Interactive Authoring and Assessment of Data-Driven Narratives
No full textData-driven narratives, which combine visualization and text to communicate quantitative insights, play an increasingly important role in how the public understands complex issues. As these narratives become more widespread, ensuring their clarity, accuracy, and trustworthiness remains a significant challenge, especially in high-stakes domains such as journalism, public health, and politics. These challenges are compounded by evolving newsroom workflows, fragmented authoring tools, and the rapid rise of generative AI.
This dissertation addresses these concerns through a combination of ecosystem analysis, empirical investigation, and interactive system design. It begins with a study of real-world data journalism practices, identifying mismatches between the needs of practitioners and the ways computational tools are typically designed and evaluated in visualization and HCI research. Building on this foundation, I curate and analyze a diverse set of problematic data narratives to develop a multi-dimensional taxonomy of common issues, which are then mapped onto a structured data communication pipeline. To support improved practices, I present two interactive systems. The first, DataWeaver, is an authoring tool that supports composing visualization and text through a bidirectional workflow, helping authors maintain alignment and accuracy during narrative construction. The second, Aletheia, supports fact-checking by connecting data claims to structured evidence using large language models, paired with interactive explanations to guide verification.
Together, these contributions integrate conceptual insight, human-centered design, technological innovation, and empirical evaluation to promote more transparent, accurate, and responsible data communication.Ph.D.Human – Centered Computin
Advanced Techno-Economic and Sustainability Analysis Methods of Emerging Grid Technologies
No full textThis dissertation develops advanced techno-economic, environmental, and asset-management methods to evaluate and guide the adoption of emerging grid technologies. A modular techno-economic analysis framework extends conventional device-level assessments by incorporating power-system simulation, with case studies showing consistent cost and performance benefits of medium voltage DC architectures. A life-cycle assessment procedure quantifies cradle-to-grave emissions, demonstrating that supercritical CO2–based circuit breakers can substantially reduce environmental impact relative to SF6 ones. To inform optimal upgrade timing, two asset-management frameworks are introduced: a two-stage framework that integrates life-cycle cost analysis with system-level optimization, and an condition-aware approach using a multi-action RMAB formulation for managing large fleets under uncertainties. Together, these methods provide a valuable decision-support toolkit for identifying cost-effective, low-emission, and scalable pathways for integrating emerging technologies into future grids
Evaluating Equity and Access in Higher Education: Regression Discontinuity, Meta-Analysis, and Institutional-level Analysis
No full textThis dissertation evaluates equity and access in higher education through three distinct empirical studies: a regression discontinuity design, a meta-analysis, and an institutional-level regression analysis. First, the dissertation employs a Fuzzy Regression Discontinuity design to estimate the causal effects of Georgia State University’s Summer Success Academy (SSA) on conditionally admitted students. This study finds a statistically significant positive effect on student persistence, particularly for Pell Grant recipients. The analysis does not find a corresponding positive effect on cumulative GPA. Second, a meta-analysis synthesizes the existing evidence on summer bridge programs (SBPs) for at-risk students. This study finds a modest, statistically significant positive association with first-year persistence. This finding, however, must be interpreted with caution, given significant heterogeneity, a high risk of bias, and potential publication bias in the underlying literature. Third, an institutional-level OLS regression examines the association between institutional racial/ethnic diversity and social capital formation. This study finds that institutional context moderates diversity's baseline negative association with the formation of cross-class social ties, which becomes attenuated or positive in institutions with greater financial resources or a larger low-income, Pell-recipient student population. Collectively, this research contributes to the understanding of equity and access by demonstrating that support interventions show measurable impacts on persistence, while institutional context is critical in moderating the social outcomes of diversity
Computational Methods for Fast and Secure Distributed Optimal Power Flow
No full textDistributed optimization algorithms may be used to coordinate large numbers of distributed energy resources (DERs) across transmission and distribution networks in the future power grid. These distributed algorithms have the potential to preserve privacy and autonomy while coordinating interconnected systems, and to improve scalability for large problems. However, there are challenges that must be addressed before implementing distributed algorithms for real-world system operation. First, distributed optimal power flow (OPF) requires repeated communication between controllers, which creates vulnerability to cyberattacks. Second, distributed OPF often takes many iterations to converge for large-scale power systems. Third, many traditional distributed optimization algorithms lack convergence guarantees for mixed-integer problems, which makes it difficult to solve problems with discrete decisions in a distributed manner.
This dissertation explores methods to improve data security and reduce computation time for distributed OPF. First, this dissertation presents a machine learning-based method to detect and mitigate data integrity attacks on information shared between controllers. Second, it explores the impact of convergence tolerance, and introduces a bound tightening algorithm to prevent constraint violations at the distributed OPF operating point for a given convergence tolerance. Third, the dissertation reduces computation time by securely computing a distributed OPF warm start using privacy-preserving neural networks. The secure warm start can be combined with bound tightening to drastically reduce the number of iterations required to converge, thus reducing computation time and improving data security. Fourth, the dissertation explores making large-scale integrated transmission-distribution switching problems more tractable. To this end, the dissertation proposes developing reduced network models to be used with hierarchical optimization methods, and evaluates performance on realistic, large-scale synthetic distribution networks
Design and Integration of Embedded, Geometrically Optimized Multi-Channel Filter Banks for Miniaturized RF Modules in IoT and Wearable Applications
No full textThe rapid growth of wireless technologies for 5G/6G and the Internet of Things (IoT) has intensified the demand for compact, high-performance, and scalable radio-frequency (RF) front-end modules. Traditional surface-mount filters occupy large footprints, increase assembly complexity, and are poorly suited to multi-channel architectures for wearable and portable devices. This thesis addresses these limitations by introducing a systematic methodology for designing and integrating embedded, geometrically optimized, multi-channel filter banks within multilayer printed circuit boards (PCBs). The approach combines analytical synthesis, via optimization, geometric miniaturization, and full-wave electromagnetic validation, enabling efficient vertical stacking and co-design of RF filters. The methodology is experimentally validated through fabricated prototypes, including two vertically stacked embedded band-pass filters at 6 GHz and 12 GHz, and a four-channel low-pass filter bank with cutoffs at 7.2, 11.1, 13.1, and 16.5 GHz. Measurements confirm return loss better than 15 dB and insertion loss below 1 dB, while achieving over a tenfold reduction in footprint compared to discrete SMT solutions. These results demonstrate that multilayer PCB-embedded filters offer a practical and scalable pathway toward miniaturized RF modules, with implications for spectrum-agile IoT, wearable, and next-generation communication systems
Object Segmentation Reasoning
No full textCurrent image segmentation methods often lack flexibility, requiring retraining for new objects and struggling to interpret nuanced user requests expressed in natural language. This project addresses these limitations by exploring the integration of reasoning capabilities, via Vision-Language Models (VLMs), into the segmentation pipeline. The primary objective is to create a system capable of segmenting objects based on descriptive, free-form text prompts, thereby improving usability and adaptability for applications such as infrastructure inventory.
The research investigates two distinct architectural paradigms. The first approach, a Multi-Module Architecture, combines a reasoning VLM (Qwen) with a specialized segmentation decoder (SAM) via a trainable adapter. This method demonstrates high robustness, achieving state-of-the-art performance on reasoning benchmarks by effectively leveraging the pre-trained strengths of both components. However, analysis reveals that this disjoint architecture imposes an information bottleneck and limits end-to-end optimization.
To overcome these structural limitations, the project pivots to a second, novel approach: End-to-End VLM Segmentation. This method fine-tunes a VLM to directly generate structured geometric outputs (polygons) from text prompts, unifying reasoning and localization into a single differentiable process. A two-stage training pipeline was developed, utilizing Supervised Fine-Tuning (SFT) followed by Reinforcement Learning via Group Relative Policy Optimization (GRPO) to optimize spatial precision.
The thesis concludes with a critical comparison of these methodologies. While the Multi-Module architecture currently offers superior stability and immediate deployment utility, the End-to-End approach demonstrates a significantly higher theoretical ceiling and architectural simplicity. Although currently constrained by training instability and topological limitations of polygon representations, the End-to-End model represents the future of generalist segmentation. Key recommendations prioritize stabilizing the GRPO training process, scaling complex reasoning datasets, and investigating alternative output representations, such as low-resolution dense masks, to unlock the full potential of this unified approach