114333 research outputs found
Sort by
The roles of REL2 mediated transcriptional co-repression in maize immunity
No full textPlant pathogens are devastating plant stressors, resulting in billions of dollars of yield loss every year. Plants employ natural defense mechanisms to combat disease. Protein acetylation is a major post-translational modification affecting many cellular processes. However, how protein acetylation modulates host resistance to pathogen infection is not understood. Cochliobolus carbonum (Northern Leaf Spot) produces the effector molecule HC Toxin, a histone deacetylase inhibitor required for virulence. RAMOSA1 ENHANCER LOCUS2 (REL2) is a transcriptional corepressor homologous to TOPLESS (TPL) in Arabidopsis. TPL family members are essential in a range of biological processes, including auxin signaling and immunity. This research tests the hypothesis that protein acetylation impacts immune signaling in maize and requires REL2. Specifically, this research will 1) elucidate how REL2 acetylation state impacts maize immunity and 2) determine REL2-associated gene expression. This project uncovers novel mechanisms underpinning plant immunity in a key agricultural crop by leveraging large-scale “omics”, genetic, and molecular approaches, and allows us to construct a model of REL2 transcriptional regulation during pathogen
External Drivers of Apparel Repurposing: A Multilevel Analysis Linking Skills, Motivations, and Repurposing Pathways
No full textTextile and apparel waste is a growing sustainability challenge, with millions of tons discarded annually and limited infrastructure for reuse or recycling. Addressing this issue requires a systems-oriented approach that integrates consumer behavior into sustainable materials production and use systems. Repurposing, the process of reclaiming garments or textiles for new purposes, extends product lifecycles, reduces reliance on landfilling, and fosters consumer creativity within circular economy transitions. Guided by a VBN-informed sustainability lens and complementary creative and economic motivation perspectives, this study examines how research-grounded external drivers shape consumer participation across four repurposing levels, offering insights to inform targeted interventions that strengthen engagement in circular clothing practices. An online survey was conducted with U.S. female consumers from Generations X and Y (n = 331), recruited via Qualtrics Panels. Measures included sewing proficiency, motivational drivers, and frequency of repurposing practices. Regression analyses showed that artistic expression consistently predicted engagement at all levels. Environmental concerns strongly predicted lower-level practices, but its influence faded at advanced levels, where sewing experience became the primary driver of participation. Monetary incentives motivated restyling and additive repurposing, while sewing experience emerged as the strongest predictor for additive and intentional patternmaking. Findings demonstrate that repurposing functions both as a creative outlet and as a sustainability strategy embedded within material-use systems. By highlighting how skills, motivations, and generational context interact across repurposing levels, this research identifies leverage points for interventions that can expand consumer participation in circular practices and advance sustainable product lifecycles.This article is published as Eike R, Hashemian B, Burton M, Hustvedt G, Cho S. External Drivers of Apparel Repurposing: A Multilevel Analysis Linking Skills, Motivations, and Repurposing Pathways. J Sustain Res. 2026;8(1):e260017. https://doi.org/10.20900/jsr20260017
Unified confusion-derived learning framework for image classification
No full textThe performance of supervised deep learning image classifiers has advanced considerably due to the availability of large-scale labeled datasets and increased computational resources. However, acquiring large, labeled image datasets in specialized domains like medical imaging remains both costly and logistically challenging. This dissertation addresses the fundamental challenge of enhancing model performance under limited labeled data conditions by leveraging confusion across various phases of developing deep learning models - before, during, and after training. Confusion is defined as the condition in which an image of one class is incorrectly predicted as belonging to another class.
The first major contribution of this dissertation is the Synthesized Image Training Technique (SIT2), a novel confusion-based training framework that systematically harnesses inter-class confusion to improve model robustness. Specifically, SIT2 identifies pairs of classes that exhibit high confusion and synthesizes “not-sure" images from these pairs, thereby incorporating confusion directly during the training process. A not-sure image is defined as a synthetically generated image that incorporates features from two distinct classes, created by blending or combining samples from highly confused class pairs in approximately equal proportions. The purpose of generating these images is to embed controlled ambiguity, ensuring that the model does not prematurely converge on a single class assignment. A prediction made with excessive confidence toward one class results in the exclusion of features indicative of alternative classes that may coexist within the image. Consequently, it is crucial that the model be encouraged to attend to and preserve features from all relevant classes to maintain a more comprehensive representation. We develop three new training strategies utilizing these synthesized images: (1) the not-sure training strategy that pretrains a model using not-sure images and the original training images, (2) the sure-or-not strategy that pretrains with synthesized sure or not-sure images, and (3) the multi-label strategy that pretrains with synthesized images but predicts the original class(es) of the synthesized images. Extensive evaluation on five medical and non-medical datasets demonstrates statistically significant performance gains, with improvements of up to 7.8% accuracy on certain datasets.
The second contribution of this dissertation is ActiveConfusion, an efficient cold-start active learning framework that leverages pretext task confusion to identify the most informative samples for labeling, before model training. By exploiting confusion patterns derived from self-supervised pretext tasks, ActiveConfusion addresses the cold-start problem in active learning, a scenario in which no labeled data are initially available. Experimental results show that ActiveConfusion matches or surpasses state-of-the-art cold-start methods while reducing pretext task training time by up to 13X. Across five public datasets covering both medical and non-medical domains, the method achieves accuracy improvements of up to 11.8% on balanced data and up to 11.4% on imbalanced medical data.
The third contribution addresses the problem of limited access to specialized Large Multimodal Models (LMMs), conceptualized here as an accessibility gap. This gap is defined as the lack of access to specialized models arising from factors such as high computational costs, restrictive licensing policies, or the limited availability of domain-specific resources. To mitigate this challenge, this work proposes BIRD - Binary Inference & Resolution for Decisions, a framework for reducing the complexity of tasks assigned to general-purpose LMMs after model training. Specifically, BIRD reformulates multi-class classification problems into a series of n-versus-rest binary subproblems, so that the model only needs to distinguish between two outcomes at a time. This decomposition is expected to reduce confusion during initial predictions, since the model is not required to simultaneously discriminate among many competing classes.
Overall, this dissertation advances data-efficient deep learning by introducing novel strategies that improve model performance under limited labeled data conditions at multiple stages of deep learning model development. The proposed contributions (SIT2, ActiveConfusion, and BIRD) demonstrate effectiveness across diverse domains, including medical imaging, general computer vision, and multimodal learning. More broadly, the confusion-based paradigm introduced in this work establishes a new perspective for designing methods that reduce errors, enhance generalization, and expand the accessibility of deep learning in resource-constrained settings at various phases of model development
Neural Geometry for PDEs: Regularity, Stability, and Convergence Guarantees
No full textImplicit Neural Representations (INRs) have emerged as a powerful tool for geometric representation, yet their suitability for physics-based simulation remains underexplored. While metrics like Hausdorff distance quantify surface reconstruction quality, they fail to capture the geometric regularity required for provable numerical performance. This work establishes a unified theoretical framework connecting INR training errors to Partial Differential Equation (PDE) (specifically, linear elliptic equation) solution accuracy. We define the minimal geometric regularity required for INRs to support well-posed boundary value problems and derive \emph{a priori} error estimates linking the neural network's function approximation error to the finite element discretization error. Our analysis reveals that to match the convergence rate of linear finite elements, the INR training loss must scale quadratically relative to the mesh size.This is a preprint from Karki, Samundra, Adarsh Krishnamurthy, and Baskar Ganapathysubramanian. "Neural Geometry for PDEs: Regularity, Stability, and Convergence Guarantees." arXiv preprint arXiv:2602.02271 (2026). doi: https://doi.org/10.48550/arXiv.2602.02271
State-of-the-art review on mechanisms and microstructural insights of eggshell powder for clay soil stabilization
No full textInnovative, cost-effective, and sustainable soil stabilization approaches are increasingly needed to address the challenges posed by problematic clay soils in geotechnical engineering. Eggshell waste, generated globally in millions of tons each year, has emerged as a promising bio-based stabilizing agent due to its high calcium content and cementitious potential. Although numerous experimental studies have reported improvements in strength and index properties of clay soils treated with eggshell powder (ESP), existing research remains fragmented, with limited synthesis of stabilization mechanisms and insufficient linkage between microstructural evolution and macroscopic soil behavior. This state-of-the-art review critically examines the role of ESP in clay soil stabilization by integrating reported changes in key geotechnical performance indicators with microstructural evidence from advanced characterization techniques. Emphasis is placed on understanding the interactions governing clay–ESP bonding and the development of soil strength. The review indicates that ESP contents up to approximately 10 % by dry soil weight can enhance soil strength and stability; however, effectiveness varies significantly with soil mineralogy, ESP processing methods, and curing conditions. This review also identifies critical research gaps, including the lack of long-term durability assessments, limited field-scale validation, and incomplete environmental and life-cycle evaluations. Addressing these gaps is essential for advancing the reliable and sustainable application of ESP in geotechnical engineering practice.This article is published as Hasheminezhad, Araz, Bo Yang, Mohammad Ahmad Alsheyab, Zexi Yin, Halil Ceylan, and Sunghwan Kim. "State-of-the-art review on mechanisms and microstructural insights of eggshell powder for clay soil stabilization." Next Materials 11 (2026): 101658. doi: https://doi.org/10.1016/j.nxmate.2026.101658.The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors gratefully acknowledge Iowa Department of Transportation and Iowa Highway Research Board for supporting this study (Grant Number: TR- 810)
Towards runtime safety assurance of autonomous cyber-physical systems through runtime verification
No full text Autonomous cyber-physical systems are often considered safety-critical due to their close interactions with humans within their environments; therefore, these systems must uphold safety requirements during deployment. While techniques such as system testing and applying formal methods (e.g., model checking and theorem proving) are often utilized to increase trust in these systems, there is a chance that a corner case was missed due to the restrictions of these techniques and the complexity of autonomous cyber-physical systems and their environments. Hence, we focus on applying the semi-formal technique of runtime verification (RV) to also enable fault detection in real-time. In this dissertation, we focus on advancing three key areas of RV: (1) enabling effective fault detection, (2) improving the responsiveness, realizability, and correctness of RV monitors, and (3) lowering barriers to increase adoption of RV monitors. To demonstrate effective fault detection, we present a case study deploying the R2U2 RV monitor on a CubeSat to detect faults and trigger appropriate mitigation actions for its Electrical Power System. Additionally, mitigation actions often have an associated deadline (e.g., if it takes a vehicle three seconds to come to a complete stop, then the vehicle must apply the brakes three seconds before a complete stop is required to mitigate an impending crash); therefore, effective fault detection may need to predict future faults. Consequently, we develop Model Predictive Runtime Verification (MPRV) to allow for the detection of future faults by predetermined deadlines. Since autonomous cyber-physical systems operate in complex environments, there may be several future behavior modes that need to be considered (e.g., a vehicle may stop, slow down, speed up, turn, etc.); hence, we also extend MPRV to Multimodal MPRV (MMPRV) to allow reasoning over several different behaviors. We also investigate how to increase the responsiveness and realizability of RV monitors to increase the feasibility of deployment on resource-constrained systems while also ensuring correct fault detection. The R2U2 runtime monitoring framework already provides real-time guarantees and a resource-aware architecture, but to increase the responsiveness, realizability, and correctness of R2U2, we manually transpile R2U2's previous C implementation to safe embedded Rust, reduce the resource overhead of both R2U2's C and Rust realizations through various optimizations, and provide guarantees of correctness through hand-constructed proofs, testing, and deductive code verification. Lastly, RV monitors are only effective if they are actually utilized, yet there are several barriers that limit the use of RV monitors, e.g., learning barrier, skepticism, challenge of formalizing system requirements. As a result, we develop the R2U2 Playground, which provides an interactive playground that provides visualization of how R2U2 evaluates specifications, decreasing the learning curve for utilizing R2U2 and enabling easier specification understanding and debugging. We also extend NASA's Formal Requirements Elicitation Tool (FRET) to configure R2U2 RV monitors from structured natural language requirements, easing the challenge of formalizing specifications from system requirements that are commonly expressed in ambiguous natural language
Graph Neural Network Architectures for interpretable I/O bottleneck analysis in high-performance computing systems
No full textHigh-performance computing (HPC) I/O performance optimization faces critical scalability
challenges. Existing automated diagnosis methods treat jobs independently, ignoring structural
dependencies between workloads with similar access patterns. Manual expert analysis cannot
scale to millions of jobs executed on production supercomputers.
This thesis presents a graph neural network approach for HPC I/O performance prediction
and bottleneck diagnosis. The method constructs k-nearest neighbor similarity graphs from
Darshan I/O profiling logs, where nodes represent jobs and edges encode behavioral similarity
based on 45 POSIX I/O features. Graph Attention Networks (GAT) and GraphSAGE leverage
this structure to learn from neighborhood information during prediction. A multi-method
interpretability framework combining attention mechanisms, GNNExplainer, and Integrated
Gradients provides robust bottleneck diagnosis through convergent evidence.
Evaluation on one million Darshan logs from NERSC Cori spanning 40 months demonstrates
4.2% improved prediction accuracy (RMSE 0.2521, R2 0.9342) over state-of-the-art ensemble
methods, with statistical significance (p < 0.05). Validation on IOR and IO500 benchmarks
confirms correct identification of known bottlenecks including small operations, alignment issues,
and filesystem configuration problems. Case study on the E2E climate model achieved 3.4×
performance improvement through optimized data decomposition and Lustre striping based on
automated diagnosis. Results establish graph-based modeling as an effective advancement for
production HPC I/O optimization
The U.S. immigrant paradox: Desistance from delinquency facilitated by Latinx youths’ preference for parental preferred language
No full textRooted in the U.S. immigrant paradox and shared language erosion theories, the current study examined whether Latinx youths’ preference for parental preferred language moderated the prediction from parental affect to relative change in youths’ delinquent behavior. Data were from a subsample of the Pathways to Desistance Study. At baseline, adjudicated Latinx youths (Mage = 16.1, N = 454, 12% female) reported on their delinquent behavior and perceived parental affect using the Self-Report of Offending scale (SRO) and Quality of Parental Relationships Inventory, respectively. Delinquent behavior was reported again using the SRO after 7 years. The moderating effect of a common language within Latinx father-offspring dyads on the prediction from paternal warmth to youth desistance was identified in the interaction tests but not in the simple slopes tests. Parental hostility predicted a relative increase in youths’ delinquency among parent-offspring dyads wherein the youth did not typically use the parent’s preferred language (i.e., English or Spanish) but not among parent-offspring dyads wherein a common language was shared. That is, among high-risk Latinx youths, a parent-youth common language showed buffering effects against parental hostility. Practical, theoretical, and methodological implications of the findings are discussed.This article is published as Chen, C.-F., Schofield, T. J., & Russell, D. W. (2026). The U.S. immigrant paradox: Desistance from delinquency facilitated by Latinx youths’ preference for parental preferred language. Current Psychology ,45 (Article 314). https://doi.org/10.1007/s12144-025-08755-2Funding: Open access funding provided by Academia Sinica. This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Data: Details about the Research on Pathways to Desistance data collection procedures and instruments are publicly available at https://www.icpsr.umich.edu/web/ICPSR/studies/2996
Investigating biophilic design features in university library study spaces
No full textThis thesis investigates how integrating multisensory biophilic elements into individual study rooms can influence graduate students’ perceived study performance and productivity, perceived environment, perceived well-being, perceived satisfaction, and preference and sound perception. While prior research demonstrates that biophilic design can enhance comfort, attention, and emotional balance, most studies focus on classrooms, workplaces, or virtual simulations, with limited empirical work in real, small-scale university study rooms. This thesis addresses this gap through a controlled, real-environment intervention conducted in two individual study rooms at the university library, one control room, and one treatment room incorporating multisensory biophilic features. The treatment room included daylight-simulating lighting, real plants, natural wood materials, subtle nature-inspired scent, and non-rhythmic rustling-leaves audio. Twenty-two graduate students participated using a within-participant A/B exposure design, completing two consecutive 45-minute sessions followed by validated quantitative surveys measuring perceived study performance and productivity, perceived environment, perceived well-being, perceived satisfaction and preference, and sound perception. Results from descriptive statistics and t-tests showed that the treatment room consistently received higher ratings across all outcome measures for both groups. Participants reported perceived improvement in focus, environmental comfort, and a clear preference for the treatment room over the control setting. Multisensory Experience, all elements combined and daylight lamp produced the most substantial perceptual effects, while plants, wood textures, nature soundscape, and scent contributed meaningfully to overall comfort and atmosphere. Positive responses to non-rhythmic auditory cues add new evidence to an underexplored area of biophilic design research. Limitations included the small sample size, restricted room modifications, short exposure duration, and reliance on self-reported data. Recommendations for future research include larger and more diverse samples, more extended exposure periods, objective behavioral or physiological measures, testing each biophilic element individually, and evaluating semi-permanent or permanent design installations. Overall, this thesis provides empirical support for multisensory biophilic strategies in small academic study rooms. The findings demonstrate that even low-cost biophilic enhancements can strengthen students’ comfort, perceived cognitive performance, and emotional well-being, offering a practical framework for evidence-informed interior design in university library environments
Thermomechanical processing of Al-Ce-based alloys and their mechanical properties for high temperature applications
No full textAluminum is lightweight, castable, corrosion resistant, and the most abundant structural metal in the Earth’s crust, making it indispensable in automotive and aerospace applications. However, the common limitation to the application of Al alloys is their limited operating temperature range. A low melting temperature range is associated with grains and precipitate coarsening at temperatures higher than ~200oC [25]. Current commercial Al alloys exhibit rapid strength degradation above ~200 °C due to precipitate dissolution and microstructural coarsening, creating a significant gap in high-temperature alloy design. In this work, Al-Ce alloys are proposed as a promising alternative due to the in situ formation of the stable Al₁₁Ce₃ intermetallic phase in these alloys. This phase is incoherent with the Al matrix, enabling Orowan looping as a strengthening mechanism while suppressing precipitate and grain coarsening because of the negligible solubility and diffusivity of Ce in Al. As a result, Al-Ce alloys retain mechanical integrity at elevated temperatures without requiring post-solidification heat treatment.
While Al-Sc alloys are known for superior properties, their high cost and limited availability constrain widespread use. Cerium, by contrast, is an abundant and low-cost by-product of the extraction and purification of Nd, Pr, Dy and Tb. This dissertation explores compositional optimization and thermomechanical processing of Al-Ce alloys to enhance both room-temperature and high-temperature performance. For example, extrusion of as-cast Al-10Ce and Al-10Ce-4Mg alloys increased their yield strengths from 62.3 MPa and 176 MPa to 94 MPa and 200 MPa, respectively, due to microstructural refinement. Building on these results, a new multicomponent alloy, Al-10Ce-4Mg-xCu-yZr, was designed, processed, and characterized. This alloy system exhibits high property retention at both ambient and high temperatures. These findings demonstrate that systematic compositional tuning and process design can produce lightweight Al-Ce alloys with high strength retention up to 300 °C, addressing a critical limitation of current commercial alloys and highlighting their potential for next-generation aerospace and automotive application