University of Illinois Urbana-Champaign
Illinois Digital Environment for Access to Learning and Scholarship RepositoryNot a member yet
123813 research outputs found
Sort by
Enhancing large language models: toward more reliable and equitable NLP
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Rishabh Adiga, accepted the attached license on 2025-04-26 at 09:28.The student, Rishabh Adiga, submitted this Thesis for approval on 2025-04-26 at 09:34.This Thesis was approved for publication on 2025-04-28 at 18:25.DSpace SAF Submission Ingestion Package generated from Vireo submission #22007 on 2025-10-19 at 18:11:05Recent advances in large language models (LLMs) have enabled them to achieve striking fewshot performance on complex tasks, yet they can remain sensitive to prompt configurations and prone to subtle biases. This thesis addresses two central problems: (1) selecting informative few-shot examples and (2) localizing and mitigating bias in ambiguous comparative prompts. First, we propose a complexity-based approach for selecting examples in few-shot sequence tagging tasks, aiming to align test examples with training examples based on syntactic and semantic metrics. By focusing on features like sentence similarity, length matching, and label diversity, we achieve more consistent and robust outcomes without fine-tuning or adding parameters. Second, we develop a method to localize and mitigate bias in LLMs by examining the attention layers that favor certain entities over others. We then scale attention in those identified layers to reduce skewed preferences, preserving overall model fluency while mitigating biases. Extensive evaluations confirm that this targeted attention manipulation provides a lightweight way to address fairness concerns without sacrificing downstream accuracy
The development, usability, and preliminary efficacy of a vocational rehabilitation guidebook
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Kathryn Huang, accepted the attached license on 2025-04-29 at 10:59.The student, Kathryn Huang, submitted this Thesis for approval on 2025-04-29 at 11:54.This Thesis was approved for publication on 2025-05-06 at 16:19.DSpace SAF Submission Ingestion Package generated from Vireo submission #22072 on 2025-10-19 at 18:11:15Background: Students with mobility disabilities may perceive barriers in accessing post-secondary education for several reasons, one of which is financial support. Parents of students with disabilities are typically the primary support for their children in applying for services to support their education but may lack self-efficacy and literacy in how to support them post-high school. Objective: This study describes the development of a guidebook intervention paired with a seminar designed specifically for parents of students with mobility disabilities who are interested in post-secondary education and examines the preliminary efficacy of the intervention. This study aimed to educate parents of students with mobility disabilities on the processes of applying for Supplemental Security Income (SSI) and Vocational Rehabilitation Services (VRS). Methods: This pre/post-intervention, mixed-methods study included parents of students with mobility disabilities aged 12-23 who are interested in attending post-secondary education. Participants engaged in an educational intervention program that included a seminar and guidebook. Prior to the start of the seminar, participants completed a pre-intervention survey evaluating their confidence and knowledge of SSI and VRS. After the seminar, participants reviewed the guidebook and completed a post-intervention survey assessing the guidebook's usability and their confidence and knowledge of SSI and VRS. In addition, 20% of participants were randomly selected for follow-up interviews. Participants: Forty parents completed the survey, seven completed the interview. The median age of their children was 16, 29 of the students had a congenital disability, 11 acquired their disability. Five families were previously connected to Vocational Rehabilitation, 35 were not. Results: Statistically significant increases in confidence (p < 0.00) and knowledge (p < 0.00), were seen after exposure to the intervention. High usability (M = 9) scores were also observed. Qualitative results further expanded participants’ views on the intervention and indicated that they valued the development and use of the intervention. Conclusion: This study is the first to describe the development of an intervention to support parents of children with mobility disabilities to access SSI and VRS services, and examine the preliminary efficacy of the intervention. Further efforts are needed to distribute the guidebook, examine the impact on enrollment among students with disabilities, and examine the impact of the intervention among a wider population of persons with disabilities
Adaptive runtime techniques for node-aware resource optimization
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Kavitha Chandrasekar, accepted the attached license on 2025-04-30 at 08:47.The student, Kavitha Chandrasekar, submitted this Dissertation for approval on 2025-04-30 at 09:02.This Dissertation was approved for publication on 2025-04-30 at 15:22.DSpace SAF Submission Ingestion Package generated from Vireo submission #22107 on 2025-10-19 at 18:11:22Adaptive runtime systems have to deal with an increasing number of cores per node. Adaptive algorithms that worked well when the number of cores per node was smaller now run into many challenges. This dissertation is aimed at addressing these challenges along multiple frontiers. The first challenge is simply about how many cores on each node to utilize for a given application. We analyzed this issue from the point of view of performance as well as energy saving, and demonstrate adaptive runtime techniques for tuning the number of cores used during execution. Secondly, many of the issues discussed here are meaningful mainly when one is using a shared-memory process spanning multiple cores of a node, rather than a "Charm++ everywhere" (analogous to MPI everywhere) mode, where each process runs on a core of its own. The latter is fundamentally sub-optimal, because it precludes resource and data-structure sharing. However, the former suffers from performance issues, especially in communication, because it is harder to expose and exploit parallelism in the network interface usage. We address this issue with comprehensive analysis and communication layer design driven by such analysis to demonstrate on-par communication performance and/or identify additional issues that need to be dealt with. Thirdly, adaptive runtime systems often have to deal with fine-grained communication. This necessitates adaptive aggregation of many short messages. However, large multicore nodes complicate this aggregation functionality, along with the diverse needs of applications. We develop metrics for characterizing the use-cases, and develop adaptive algorithms that work well on multicore nodes and for varied use cases. Fourthly, dynamic load balancing is another challenge: migrating work units to cores to effectively balance load in dynamic applications needs to take into account the multi-level nature of communication, while also addressing scalability issues created by the huge number of cores on modern machines. We demonstrate node aware, and communication aware strategies, including refinement strategies that aim at minimal migration and low load balancing overhead
A data-driven method for improving a black-box controller
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Alex Gisi, accepted the attached license on 2025-05-02 at 13:50.The student, Alex Gisi, submitted this Thesis for approval on 2025-05-02 at 13:55.This Thesis was approved for publication on 2025-05-05 at 12:32.DSpace SAF Submission Ingestion Package generated from Vireo submission #22184 on 2025-10-19 at 18:11:37Optimal control design is an important engineering task. The optimality of a controller is measured by how well closed-loop system trajectories under that controller satisfy a given measure of performance. If a controller is sub-optimal with respect to the performance measure, it would be beneficial to re-design it; however, controller re-design can be expensive. Therefore, it is desirable to alter the closed-loop system performance without touching the baseline controller, that is, by treating it as a black-box. This thesis proposes a method for doing so based on data collected by observing closed-loop trajectories under the baseline controller. The method is based on gain scheduling, or multiplicative modulation of the baseline control signal. A gain scheduling policy describes how and when to apply gains to alter the baseline signal. A gain scheduling policy parameterization and training algorithm for automatically improving the black-box baseline controller is proposed. It is shown that for the proposed policy parameterization, the training can be made more efficient by applying an alternating optimization technique. The resulting gain scheduling policy and training algorithm were applied to three control systems with distinct qualities: a linear time-invariant system, an inverted pendulum, and a skid-steer mobile robot simulation. Additionally, a combined powertrain and kinematic model is developed to implement the mobile robot simulation. To perform realistic evaluations, both linear-quadratic regulator and reinforcement learning-trained controllers are used in the method evaluation. Furthermore, two other learning algorithms beside the one proposed are used to give the results context. It is shown that the proposed method can effectively improve the output of either baseline controller, although certain situations cause worsened performance. The practical implications of these results are examined using examples
From Hindu villain to heroic Buddhist king: Dathagiri in the Thiri Rama
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Hpone Tu, accepted the attached license on 2025-05-04 at 08:22.The student, Hpone Tu, submitted this Thesis for approval on 2025-05-04 at 08:31.This Thesis was approved for publication on 2025-05-07 at 08:26.DSpace SAF Submission Ingestion Package generated from Vireo submission #22199 on 2025-10-19 at 18:12:29This paper analyzes the character of Dathagiri in the Thiri Rama, an 18th century Myanmar court drama. The Thiri Rama is the Myanmar version of the Rāmāyaṇa story – this play is a little-known work in Western scholarship. The most widely known version of the Rāmāyaṇa is Vālmīki’s Rāmāyaṇa, and in this story, the villain is Daśagrīvā. In the Thiri Rama, however, I argue that Dathagiri, Daśagrīvā’s counterpart, is not a villain, but a heroic Buddhist figure. Scholars of the Thiri Rama have labelled Dathagiri the villain of the story, taking, I suspect, a reading of Daśagrīvā and imposing it upon Dathagiri. But this claim does not withstand close scrutiny. I textually analyze the first few acts of the play—which contain Dathagiri’s coming-of-age story—to demonstrate that Dathagiri is a heroic, Buddhist king. Furthermore, to truly understand the complexity of the Dathagiri character, I make connections to the historical context of the Thiri Rama. The Dathagiri character also contains in it the voices of Siamese captives after the fall of Ayutthaya in 1767 and that of the Myanmar royal court defending its empire against a Qing invasion. My study of this relatively obscure text, I hope, enriches the scholarship on Buddhist kingship and the Rāmāyaṇa
Enhancing safety and resilience in AI-driven systems: from autonomous vehicles to data centers
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Ziheng Chen, accepted the attached license on 2025-05-07 at 16:54.The student, Ziheng Chen, submitted this Thesis for approval on 2025-05-07 at 17:09.This Thesis was approved for publication on 2025-05-08 at 14:04.DSpace SAF Submission Ingestion Package generated from Vireo submission #22258 on 2025-10-19 at 18:13:49Artificial intelligence (AI) has been widely adopted due to its advanced capabilities, finding applications in autonomous vehicles and, more recently, in large language models (LLMs) powered by transformers. As AI systems continue to expand in scope and influence, ensuring their reliability and robustness becomes increasingly critical, particularly in unforeseen scenarios where failures can have significant consequences. Furthermore, given the substantial computational and economic costs associated with training and deploying large-scale models, which often require extensive GPU clusters, the underlying infrastructure must be highly fault-tolerant to maintain efficiency and stability. This thesis presents our work on iPrism, a framework designed to enhance the safety of autonomous vehicles by reducing collision rates through AI-driven risk assessment and mitigation. This thesis also discusses the work Characterizing GPU Resilience and Impact on AI/HPC Systems, which examines GPU failure patterns in the Delta Supercomputer to improve system reliability
Towards autonomous excavation in tough terrains and unobservable obstacles
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Noah Franceschini, accepted the attached license on 2025-05-08 at 16:33.The student, Noah Franceschini, submitted this Thesis for approval on 2025-05-08 at 16:42.This Thesis was approved for publication on 2025-05-09 at 10:32.DSpace SAF Submission Ingestion Package generated from Vireo submission #22272 on 2025-10-19 at 18:13:58Current autonomous excavation methods often make the assumption that the terrain within their environment is homogeneous, and remains free of obstacles. This assumption is insufficient for safe and reliable systems in the real world. Real autonomous excavation agents have to adapt to uncertainty within their environments such as changing terrain conditions and colliding with obstacles during an excavation fill cycle. In this work, we seek to improve autonomous excavation methods by introducing a novel set of scooping primitives to assist in breaking up jammed materials, a novel Reactive Attractor Impedance Controller (RAIC) that dampens trajectories when facing high impedances during the scooping cycle, and finally, we introducing a framework for material parameter estimation using tactile feedback. Our most novel methods comes in the form of our scooping primitives and our RAIC controller. These primitives are a set of oscillatory movement primitives with the intention of breaking up jammed particles encountered during the robot’s excavation cycle. Our RAIC controller works in tandem with our scooping primitives to help dampen and stop the robot’s end-effector trajectory when experiencing high forces to then break up the jammed materials. We evaluated these methods on a variety terrains and found them to vastly improve material scooped, as well as ensuring better safety over standard methods. Our material parameter estimation framework proposes using a simple force estimation method when dragging and intruding surface through granular media. This method, Resistive Force Theory (RFT), allows us to predict forces on our end-effector, while only requiring a single parameter ζ, the terrain’s resistivity coefficient. In addition to this RFT model, we propose a material deformation model based on the estimated forces to displace and settle the granular media between time steps. We then go on to explain how, in future works, this model can be used to estimate the local ζ to create a heterogeneous terrain property map to be used in excavation environments
Over-parameterized low-rank matrix estimation: Theory, algorithms, applications
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Jialun Zhang, accepted the attached license on 2025-04-14 at 01:48.The student, Jialun Zhang, submitted this Dissertation for approval on 2025-04-14 at 01:53.This Dissertation was approved for publication on 2025-04-20 at 16:06.DSpace SAF Submission Ingestion Package generated from Vireo submission #21599 on 2025-10-19 at 18:17:12This thesis considers the non-convex methods for low-rank matrix estimation based on Burer-Monteiro factorization, which offers better scalability and lower per-iteration costs compared to convex methods. However, non-convex optimization faces challenges from spurious local minima or critical points, which can lead gradient-based algorithms to converge to suboptimal solutions. Prior research has shown that a sufficiently large training dataset can eliminate spurious minima in non-convex problems, enabling successful low-rank estimation. However, these guarantees are often impractical in real-world applications for two reasons. First, they require a large amount of data to remove spurious local minima across the entire parameter space, which is costly and challenging for large-scale applications. Second, they assume ``regular" measurement conditions (e.g., restricted isometry property or incoherence), which are rarely satisfied in practice. To address these challenges, we first prove that a good initial point can reduce the required sample size to avoid spurious minima near the ground truth. This insight reveals that practical low-rank estimation problems are often less demanding in terms of data than prior theory suggests. Specifically, there is a direct tradeoff: better initial point quality linearly reduces sample complexity, and vice versa. The thesis thus sheds light on how non-convex landscapes become more manageable with improved initialization or increased samples, providing insights that could extend to other non-convex problems. The major focus of our thesis is how over-parameterization can be used to effectively solve non-convex matrix estimation problems. Over-parameterization serves two purposes: it certifies global optimality by allowing sufficient flexibility in the model, and it ensures solution accuracy even when the true rank is hard to estimate. Unfortunately, over-parameterization slows gradient descent exponentially, which dramatically increases the cost to converge to the ground truth. To overcome this, the thesis introduces PrecGD (preconditioned gradient descent), a method that preserves the fast convergence of gradient descent, achieving exponential error reduction per iteration. PrecGD’s convergence rate is independent of ill-conditioning or over-parameterization, with a per-iteration cost only slightly higher than standard gradient descent. Finally, we also extend our algorithm, PrecGD, to two important practical settings: (1) first, we present a stochastic version of PrecGD that applies even to huge-scale matrix completion type problems where even on full-batch gradient evaluation is too expensive. (2) Then, we consider adapting our algorithm to scenarios where the measurements are extremely noisy, which makes preconditioning significantly more difficult since it is easy to magnify the noise as well. In both cases we present a rigorous proof for the success of our method, which is also validated extensively in numerical experiments
Exploring the influence of employee motivations on workplace learning effectiveness and subsequent job performance: A mixed methods study within the W department at X engineering
Get PDFSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Kuang-Pu Cheng, accepted the attached license on 2025-02-12 at 19:36.The student, Kuang-Pu Cheng, submitted this Dissertation for approval on 2025-02-12 at 19:38.This Dissertation was approved for publication on 2025-03-11 at 13:16.DSpace SAF Submission Ingestion Package generated from Vireo submission #21642 on 2025-10-19 at 18:17:28This study explores the influence of employee motivations on workplace learning effectiveness and subsequent job performance within the W Department at X Engineering. It integrates human resource development (HRD) theories with empirical research. Specifically, the study is grounded in self-determination theory (SDT), which provides a framework for understanding the motivational dynamics that influence workplace learning effectiveness and subsequent job performance in organizational settings. The study was conducted using an explanatory sequential mixed-methods design. It began with a quantitative survey of the W Department’s employees. This survey assessed the influence of intrinsic and extrinsic motivations, social and environmental factors, and basic psychological needs on workplace learning effectiveness and subsequent job performance. This was followed by qualitative interviews to enrich the quantitative data, offering insights on how these factors interact within the specific context of the W Department at X Engineering. Quantitative findings reveal that the social and environmental factors, the organization development (OD) interventions in this study, is the primary influencer for both workplace learning effectiveness and subsequent job performance, with intrinsic motivation also showing significant impacts. This element is essential for fostering a supportive learning environment and promoting active engagement in professional growth. The qualitative insights emphasize that while intrinsic motivation deeply resonates with individual employees, enhancing their engagement and satisfaction, the social and environmental factors also plays a crucial role in influencing their learning experiences and job performance. Basic psychological needs, particularly the autonomy, is highlighted as central in the qualitative descriptions, although it does not show significance in the quantitative analysis. The implication of this study suggests the need for HRD practices that not only foster motivation but also create environments that support the thorough development of employees. It also provides additional empirical understanding of how motivational factors can be strategically aligned with HRD programs to improve workplace learning effectiveness and subsequent job performance. This alignment could enhance organizational efficiency and employee satisfaction. Moreover, future studies could benefit from exploring these motivational factors across different organizational settings to understand their long-term effects and potential applicability more broadly. Considering the limitations such as the homogeneity of the sample and its focus on a single department, expanding research efforts could enhance the generalizability of these findings within HRD practice
Stoichiometric rare-earth materials at cryogenic temperatures as a platform for quantum memory devices
No full textSubmission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo termsThe student, Donny Pearson, Jr., accepted the attached license on 2025-02-24 at 15:34.The student, Donny Pearson, Jr., submitted this Dissertation for approval on 2025-02-24 at 16:15.This Dissertation was approved for publication on 2025-03-03 at 10:01.DSpace SAF Submission Ingestion Package generated from Vireo submission #21650 on 2025-10-19 at 18:17:35Rare earth ions in solids at cryogenic temperatures are an excellent platform for developing quantum memory, a device that can store and retrieve single photons on demand. We seek to realize such a device toward the ultimate goal of long-distance quantum communication. Rare earth ions in solids have long-lived optical and spin coherence properties—milliseconds and as long as hours, respectively. Their unique electronic structure affords them these long coherence properties. Additionally, they lack motional dephasing and can have a high density of emitters. The primary challenge we face is the inhomogeneous broadening of the optical transition -- site-to-site variations of the energy levels due to defects in the solid host. The rare earth ions experience different local environments, resulting in a spread of transition frequencies that can be many orders of magnitude broader than the ion’s narrow homogeneous linewidths and the hyperfine spin state splitting. Due to the inhomogeneous broadening, we cannot optically resolve the hyperfine spin states in our ion of choice, Eu3+. The consequence of our inability to resolve the hyperfine states is that we cannot efficiently utilize the hyperfine spin states' long-lived coherence properties for an efficient quantum memory. Rare earth ions are traditionally doped within the host matrix and are themselves defects in the solid. Stoichiometric materials, materials in which the elements appear in proportion to their chemical formula, are a possible avenue toward realizing high-density rare earth materials with ultranarrow inhomogeneous linewidths. Stoichiometric materials are moving to a regime of 100% doping regime where the local environment of the rare earth ions should be more homogeneous. Stoichiometric materials offer an avenue to reduce inhomogeneity, enabling efficient quantum memory with long storage times. Here, I will discuss several materials that we have probed toward the goal of finding an environmentally stable Eu\ion{} material with ultranarrow optical linewidths for a quantum memory device. The first materials we explored were two metal-organic frameworks, EF and EFFA, where we characterized their luminescence properties at room temperature and 1.4 K. The next material we investigated was sodium europium tetrakis(iodate), which exhibits inhomogeneous linewidths as narrow as 2.2(1) GHz and a homogeneous linewidth of 120(10) kHz. Additionally, we have measured the hyperfine spin lifetime to be > 1 s, which sets the ultimate limit on the hyperfine spin coherence time and, thus, the maximum storage time. Finally, we implement a preliminary demonstration of an atomic frequency comb, pushing forward our goal of utilizing sodium europium tetrakis(iodate) as a quantum memory. Preliminary results for other materials I have explored are also included, such as: europium trialuminum borate, europium tris(iodate) and bis(tetramethylammonium) potassium hexakis(nitrato)-europium(III)