DRUM - Digital Repository at the University of Maryland

DRUM collects, preserves, and provides public access to the scholarly output of the university. Faculty and researchers can upload research products for rapid dissemination, global visibility and impact, and long-term preservation.

Submit to DRUM

To submit an item to DRUM, login using your UMD credentials. Then select the "Submit Item to DRUM" link in the navigation bar. View DRUM policies and submission guidelines.

Equitable Access Policy

The University of Maryland Equitable Access Policy provides equitable, open access to the University's research and scholarship. Faculty can learn more about what is covered by the policy and how to deposit on the policy website.

Theses and Dissertations

DRUM includes all UMD theses and dissertations from 2003 forward.

List of Communities

Collections Organized by Department

A. James Clark School of Engineering

College of Agriculture & Natural Resources

College of Arts & Humanities

College of Behavioral & Social Sciences

College of Computer, Mathematical & Natural Sciences

UM Community-managed Collections

Academy for Excellence in Teaching and Learning Student Essay Award

Alliance for Coastal Technologies

Applied Research Laboratory for Intelligence and Security (ARLIS)

Archaeology in Annapolis

Barrio de Langley Park

Recent Submissions

TOWARDS PRINCIPLED AI AGENTS WITH DECENTRALIZED AND ASYMMETRIC INFORMATION
(2026) Liu, Xiangyu; Zhang, Kaiqing; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
AI models have been increasingly deployed to develop autonomous agents for decision-making, with prominent applications in Go, video games, robotics, autonomous driving, healthcare, and human assistance. Many of these successes involve multiple AI agents interacting dynamically with one another and with humans. At the same time, these systems often face challenging partial observability, further complicated by asymmetric information between training and testing and decentralized information across agents. This thesis aims to lay the theoretical foundations for principled AI agents operating under partial observability with asymmetric and decentralized information. First, we study reinforcement learning agents in partially observable Markov decision processes when privileged information is available during training but not at test time, a common setting in robot learning and deep RL. We revisit two major empirical paradigms, expert distillation, also known as teacher-student learning, and asymmetric actor-critic, and identify their limitations in finding near-optimal policies. We then develop a principled algorithm with polynomial sample complexity and quasi-polynomial computational complexity, revealing the provable benefits of privileged information for AI agents in partially observable environments. Second, we study multi-agent reinforcement learning with information sharing under decentralized information. To circumvent known hardness results and avoid computationally intractable oracles, we advocate leveraging potential information sharing among agents. We establish several computational complexity results that justify the necessity of information sharing and appropriate observability assumptions. Motivated by the inefficiency of planning in the ground-truth model, we further approximate the shared common information to construct an approximate model of the partially observable stochastic game, in which approximate equilibrium planning can be quasi-efficient under these assumptions. We then develop a partially observable multi-agent RL algorithm that is both statistically and computationally quasi-efficient. Third, we study multi-agent RL with latent state representations under decentralized information. Beyond using common information in the raw observation space, we propose to align different agents' latent representations through a new representation learning framework, Representationally Aligned Approximate Latent Model, or \realm. We establish conditions under which latent-model equilibria exist and can be used to solve the original dynamic game before compression. We also develop provable representation learning algorithms for computing such latent-model equilibria with both computational and statistical efficiency. Along the way, we also design an efficient learning algorithm for an important special case of \realm, partially observable stochastic games with deterministic filters, which improves existing results by addressing the curse of multiagency and relaxing the required privileged-information assumptions. Finally, we examine large-language-model-powered agents, which use LLMs as the main controller for decision-making, with the goal of understanding and enhancing their capabilities in canonical decentralized and multi-agent scenarios. In particular, we use regret, a standard metric in online learning and RL, to study the in-context decision-making limits of LLM agents through controlled experiments. Motivated by the observed no-regret behaviors, we propose a hypothetical model that can well explain such behaviors and prove it is a natural consequence of pre-training via next-token-prediction. Together, these results provide a principled understanding of AI agents under partial observability, asymmetric information, and decentralized information, paving the way toward practical AI agents for real-world strategic and decentralized systems.
POST-TRAINING OF VISION-LANGUAGE AGENTS FOR DECENTRALIZED AUTONOMOUS VEHICLE COORDINATION USING GENERALIZABLE MULTI-AGENT REWARDS
(2026) Cole, John Robert; Goldstein, Thomas A; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
Decentralized coordination at unsignalized intersections remains a persistent failure mode formodern autonomous driving policies when vehicle-to-everything (V2X) communication is unavail- able. Policies trained primarily with ego-centric objectives (e.g., collision avoidance, comfort, and action consistency) can be overly conservative in symmetric interactions, leading to deadlocks, or can make conflicting commitments, leading to unsafe near-collisions. This thesis addresses this gap by introducing a social post-training method for Alpamayo-R1 (AR1) that explicitly rewards behavior that is predictable to neighboring agents. We extend AR1’s Group Relative Policy Optimization (GRPO) post-training by augmenting thereward with Expectation Alignment (ELIGN), an intrinsic social term that penalizes mismatch between a learned neighbor-expectation model and the realized shared next observation. To make ELIGN applicable to AR1’s continuous trajectory outputs, we define the shared observation space over low- dimensional kinematic waypoints (x, y, ψ, v) rather than high-dimensional perception features, and we learn a compact trajectory prediction model offline before fine-tuning. The composite reward combines a trajectory-fidelity L2 term, a comfort score, and the ELIGN social penalty under a gated formulation that prevents the social term from masking large trajectory failures. Post-training is implemented using the cosmos rl framework with ReasoningVLAGRPOTrainer and vLLM-accelerated rollout generation, representing the first application of GRPO to a production-scale VLA model in a neural- rendered closed-loop driving simulator. We evaluate the proposed AR1+ELIGN post-training in a multi-agent simulation benchmark ofsymmetric four-way arrival scenarios in AlpaSim and compare against an ego-centric AR1 baseline as well as standard multi-agent reinforcement learning baselines (PPO and MAPPO). Performance is measured by collision rate (as a hard safety constraint), deadlock rate, intersection clearance time, and jerk variance as an indicator of indecision. Finally, we study zero-shot social generalization by testing whether ELIGN-fine-tuned agents coordinate effectively with novel partner agents not encoun- tered during training. Quantitative benchmark results are pending completion of GRPO training; pre- training reward validation and closed-loop baseline evaluation confirm that the reward pipeline is stable and well-calibrated, with the social term contributing a mean penalty of approximately −0.02 per step under the deployed weighting while the trajectory-fidelity signal remains dominant.
INCLUSIVE REFUGEE EDUCATION PRAXIS: EXPLORING REFUGEE-LED EDUCATION FOR CHILDREN WITH LEARNING DIFFICULTIES
(2026) Safarha, Elnaz; Zakharia, Zeena; Klees, Steven; Education Policy, and Leadership; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In the context of forced displacement, inclusive education (IE) remains a critical yet elusive concept, particularly when refugeehood intersects with special educational needs and disabilities (SEND). Despite global frameworks recognizing and advocating for inclusive refugee education, implementation continues to fall short, most acutely for refugees with SEND. Key challenges include ambiguous definitions of inclusion (El Ahmad, 2022), insufficient funding (Crea et al., 2022), inadequate teacher training (Hadidi & Al Khateeb, 2015), and a lack of inclusive curricula responsive to refugees’ needs (Shuayb et al., 2016). Refugee children with SEND face compounded marginalization, as social narratives often depict refugees as burdens on social and economic systems (Kiwan, 2019) and portray children with SEND through a deficit lens that denies their capacity to benefit from education (IASC, 2019).Global frameworks have increasingly emphasized the integration of refugees into national education systems, particularly UNHCR 2012 and 2019. However, in practice, inclusion is often reduced to structural access, overlooking the relational and contextual dimensions that shape meaningful inclusion of refugees. These limitations are particularly evident in non-formal education (NFE) settings, especially those led by refugee communities, that frequently serve as primary access points when formal systems fail to include refugee learners. Refugee-led educational NFE initiatives play a vital role in addressing these gaps, drawing on community knowledge and social proximity to respond to learners’ needs. Yet, despite growing advocacy for shifting powers toward refugee communities through their direct engagement, these initiatives often lack financial support and recognition from the international humanitarian actors (Aden, 2025). At the same time, empirical research on how inclusion is conceptualized and enacted for refugees, particularly those with special educational needs and disabilities, remains limited, especially in the Global South, where most refugees reside. Using a critical ethnographic design grounded in decolonial feminist epistemologies, critical pedagogy, and critical refugee studies, the research positions refugee educators as knowledge producers rather than passive beneficiaries. Fieldwork was conducted between 2023 and 2025 across two schools within the organization, referred to here as the Hope Center. Data sources included immersive in-person and virtual observations, as well as ethnographic interviews in multiple formats with teachers, special educators, school leaders, and social workers. The study captures inclusive practices not only during routine schooling but also during periods of disruption, including the shift to remote learning amid Israeli strikes and school closure in Fall 2024. Findings reveal that inclusion, as defined and practiced by refugee educators in this context, is a systemic and evolving process, not a static goal. Rather than focusing solely on student access or barrier removal, inclusion is conceptualized as relational, holistic, and responsive to both learners’ and teachers’ needs. Through strong interpersonal trust, adaptive communication, and collaborative leadership, teachers exercise situated agency in ways that humanize education for children with and without SEND. This is while educators are also being supported, held accountable, and recognized as professionals with their own lived realities. By adopting a bottom-up lens, this study reframes inclusion as a dynamic, humanizing process shaped by educators’ knowledge, values, and relationships. It challenges dominant models that position inclusion as a policy mandate handed down from above and instead highlights inclusive practices already working on the ground. This study contributes to emerging alternative discourses on inclusive refugee education by offering a reimagined understanding of inclusion by centering refugee educators’ knowledge, lived realities, and shared values in shaping inclusive pedagogies.
Queer Specters of Intellectual Labor: Knowledge, Ethics, and Desire in the Poetics of Study
(2026) Johnson, Zachary; Hanhardt, Christina B; American Studies; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
In contemporary public and institutional discourse, the humanities are persistently compelled to justify their existence through appeals to utility, productivity, and ethical self-making. Yet the ideological and affective conditions that structure this relentless demand for justification remain critically underexamined. Approaching the “crisis” of the humanities from the vantage point of minoritized interdisciplines, such as gender, queer, and ethnic studies, this dissertation examines how the liberal university evaluates and disciplines intellectual labor by casting certain forms of knowledge as illegitimate, excessive, or driven by pathological desire. By defining the prevailing academic common sense as a "liberal metaintellectual consciousness," this project investigates the structural disavowals required to maintain the university's fantasy of objective, instrumental reason.Adopting an interdisciplinary framework that synthesizes continental philosophy, queer theory, black studies, and critical university studies, the project traces the ideological formations of academic value across public controversies and cultural texts. Chapter One examines the 2018 “Grievance Studies” affair as a paradigmatic scene through which minoritarian fields are cast as excessive, illegitimate, and contaminating forms of knowledge, situating the hoax within a longer culture-war genealogy that includes the 1996 Sokal Affair. Chapter Two turns from scandal to defense, arguing that contemporary justifications of the humanities often reproduce a liberal metaintellectual consciousness organized by utility, legitimacy, and moral repair, a structure this project names the “respectability politics of thought.” Chapter Three shifts from discourse to cultural form through a close reading of Candyman (1992), theorizing academic horror as a mode that exposes the racialized desire and epistemic violence embedded in the liberal will-to-know. Chapter Four develops the dissertation’s central alternative, a queer poetics of study, through Rose Red (2002), The Ring (2002), and Sinners (2025), reading horror’s figures of desire, opacity, and shared temporality as speculative resources for imagining intellectual life beyond mastery, productivity, and institutional justification. The dissertation concludes with a coda, “The Right to Study,” which reframes study as a collective claim to time, space, and resources for non-instrumental intellectual engagement. Grounded in a critique of Enlightenment liberalism and informed by theories of differential consciousness, this project argues that the value of the humanities should not solely be anchored in instrumental utility, performative efficacy or liberal narratives of moral redemption. Instead, it proposes a "queer poetics of study," an orientation that affirms the immanently meaningful, non-instrumental dimensions of intellectual life, including desire, curiosity, opacity, and shared temporality. Reframing intellectual labor through this speculative and affective lens, this study illuminates how reclaiming the unrespectable pleasures of study can challenge the neoliberal university's extractive demands, shifting the discourse away from institutional defense and toward the cultivation of a collective "right to study."
NAVIGATING HEALTH IN THE DIGITAL AGE: AN INTEGRATED THEORETICAL INVESTIGATION OF OLDER ADULTS’ ONLINE HEALTH INFORMATION-SEEKING INTENTIONS
(2026) Doh, Jiawen; Kim, Jiyoun; Communication; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
As internet access among older adults continues to increase, understanding how they engage with online health information has become increasingly important. This dissertation examines online health information-seeking (OHIS) among older adults who already use the internet, with particular attention to variation in their behaviors and intentions. Rather than focusing on access barriers, this study centers on differences in how older adults seek, evaluate, and use health information in an evolving digital environment. This research includes three main objectives. First, it examines whether meaningful differences exist in how older adults engage with online health information. Second, it develops and tests an integrated theoretical framework to explain OHIS intentions. Drawing on existing theoretical frameworks, the study synthesizes key constructs and incorporates additional constructs to better capture the influences on OHIS intentions. Third, it provides contextual insight into the digital environments in which OHIS occurs, including patterns of platform use and the types of health information sought.This dissertation employs a two-study design. A preliminary analysis using data from the Health Information National Trends Survey (HINTS 7) identifies patterns of OHIS behaviors among older adults with internet access. Building on these findings, a survey-based study tests the proposed model and examines relationships among key determinants of OHIS intentions. Key findings indicate that OHIS among older adults is influenced less by sociodemographic characteristics and more by digital capabilities and evaluative perceptions. Digital skills emerge as a key driver of OHIS behavior, while perceived usefulness and positive evaluations are the strongest predictors of OHIS intention. The results of both studies indicate a two-step process: among older adults, digital skills are necessary for participation OHIS, while perceived usefulness, positive evaluation, and trust-related perceptions ultimately determine their intention to seek health information online. By integrating multiple theoretical perspectives, this research advances understanding of OHIS among digitally connected older adults and provides practical implications for improving the accessibility and credibility of online health information.