University of Illinois at Chicago
University of Illinois at Chicago: UIC INDIGO (INtellectual property in DIGital form available online in an Open environment)Not a member yet
21439 research outputs found
Sort by
A Disability Studies and Critical Autistic Analysis of Peer Mentoring for Autistic Postsecondary Students
No full textPeer mentoring is a common support service for autistic students in postsecondary education. This study explores the current landscape of peer mentoring programs at colleges and universities in the United States and proposes a novel mentoring model, constellation mentoring, to support this population. Constellation mentoring is informed by mentoring literature, critical disability praxis, and lived experiences of autistic community members to address limitations in existing mentoring programs for autistic students.
This dissertation presents a survey study of program staff in thirty peer mentoring programs for autistic students and a focus group study with four autistic young adults who participated in constellation mentoring in the Chicago area. The dissertation also provides a guide for implementing constellation mentoring as a collaborative group process. The goal of constellation mentoring is to empower autistic people as agents of change in their own lives and their communities. Implications of this research may enhance autistic-centered mentoring for autistic students in postsecondary education and beyond, and thus, may interest educators, administrators, disability service professionals, college autism professionals, autistic self-advocates, community members, and others
Visual Computing Design for Explainable, Spatially-aware Machine Learning
No full textMany real-world machine learning applications require combining the strengths of human analysts who have specialized knowledge, with computational systems that excel at statistical analysis and large-scale data processing. This is especially true in the case of statistical modeling with spatial data, where information about the coordinates of each data item benefits from domain expert knowledge. For example, in many cancers, the location of the tumors and the specific pattern in which the disease spreads have an impact on the patient's treatment, and the specific organs nearby that are affected can impact long-term patient outcomes. However, the reliance on retrospective cohorts and the fact that organs with distinct functions are spatially adjacent can lead to many spurious findings when using traditional statistical approaches. This dissertation attempts to deal with these issues through the design of integrated visual computing and explainable machine-learning tools with spatial data, with a focus on human-centered co-design of these systems alongside clients.
The specific research question this dissertation aims is: How do we integrate spatial data into explainable visual computing + machine learning (VC+ML) systems? This involves several sub-challenges: 1) Domain characterization; 2) strategies for the design of models for spatial VC problems that consider requirements from both ML model builders and model clients in a collaborative setting; 3) modeling similarity between sets of spatial features; 4) the design visual encodings for to explain these spatial machine learning model predictions to clients; and 5) approaches to measure the effect of deploying these spatial VC+ML systems in practice. To answer this, I will detail five design studies focused on human-centered collaborative design of mixed VC + ML systems for real work problems that use spatial data, followed by general design insights for the design of these systems
Levels of Heavy Metals in Water Used in Artisanal Small-Scale Gold Mining in Western Kenya
No full textThis research examines the occupational exposure to metals in artisanal and small-scale gold mining (ASGM) environments, particularly focusing on the potential for dermal exposure to metals that impair human health at low concentrations such as arsenic (As), cadmium (Cd), chromium (Cr), mercury (Hg), manganese (Mn), and lead (Pb). These metals are of particular concern due to their ability to accumulate in the body, disrupt vital biological processes, and cause severe health issues even at low exposure levels.
The study analyzes metal concentrations in water sources used by ASGM workers, including primary and secondary basins involved in mercury and gold amalgamation, as well as pools at mining sites. Inductively coupled plasma mass spectrometry (ICP-MS), a well-established analytical technique, were used.
The findings reveal that ASGM workers face a significant risk of exposure to harmful metal levels through dermal contact with contaminated water, with additional insights from total exposure estimates derived from filter and filtrate analyses. Although the metal concentration analysis results are consistent with previous studies, this research fills a knowledge gap by emphasizing dermal exposure, a less-studied pathway, and offering context-specific insights into the occupational health risks in ASGM practices. The combination of filter and filtrate data provides a more comprehensive understanding of total metal exposure.
This research addresses the gap in the limited data on dermal exposure to metals in ASGM environments, offering valuable insights for future research and public health interventions. It highlights the need for improved monitoring and protective measures for workers, underscoring the importance of managing health risks to safeguard the well-being of individuals involved in ASGM practices
Algebraic Frameworks and Computational Experiments for Enhanced Machine Learning
No full textIn this thesis, we explore both the theory and practical considerations that come with tasks in machine learning.
First, we apply a deep model to segment and highlight cancerous tumors in 3D medical scans. The data we use is uniquely produced by technology that was pioneered by a medical imaging company called Clarix Imaging, and the model we produce is unseen in existing literature.
Second, we introduce two methods for reducing computational requirements for making predictions with neural networks, including by algebraically representing linear neural networks as a single layer, and by removing skip connections throughout training. We demonstrate these methods with computational examples on Residual Networks (ResNet) architecture.
Third, we use tropical geometry to derive an algorithm that finds all linear regions of nonlinear neural networks, which are regions where the network behaves as a linear map. Through experiments, we find that neural networks with skip connections have more linear regions than regular networks, even before training begins, and that unseen data lies in unseen linear regions more often, providing an understanding of the efficacy of skip connections. We also visualize linear regions and observe the underlying patterns of their distribution.
Lastly, we explore how an explainable artificial intelligence model called a GA^2M can be trained to fit computer-generated data with known underlying structure, providing insights into its interpretability and reliability
Semantic-Driven Multimodal Learning: Gradient-Based Methods and Applications
No full textMultimodal learning addresses the challenges of deep networks in processing complex datasets containing multiple modalities, such as effectively integrating data with different structures and learning modality fusions between texts and images. Traditional deep learning models are primarily inductive in nature and often struggle to capture the full semantic richness and implicit dependencies within such datasets. In this thesis, we present four recent papers and address these challenges by integrating and processing information from different sources or modalities of data. This integration enriches the understanding of individual data points and enhances the model's generalization ability in real-world applications.
We first introduce a novel self-supervised ontology matching (OM) method to fuse linguistic information and the structural information inherent in the ontologies. We proposed capturing multiple structural contexts encompassing both local and global interactions between concepts in the ontologies. Our experiments on the Bio-ML datasets, which are publicly available from Ontology Alignment Evaluation Initiative (OAEI) and tasks, demonstrated that our method surpassed the state-of-the-art OM systems in terms of alignment quality and inference time. This OM method establishes well-structured knowledge bases and data integration that are crucial for facilitating further studies.
In the second paper, we addressed the challenge of specifying domain or prior modal knowledge in a backpropagation-friendly manner in large-scale and noisy settings, such as with large Vision-Language Models (VLMs). We proposed a simplified alternative of combining features from pre-trained deep networks and freely available semantic explicit knowledge. To remove irrelevant explicit knowledge that does not correspond well with the images, we introduce an implicit differentiable Out-of-Distribution (OOD) detection layer. This layer addresses outlier detection by solving for fixed points of a differentiable function and using the last iterate of fixed point solver to backpropagate. We pre-trained on three public datasets in the experiments, including COCO, Visual Genome, and SBU Captions. We used all public datasets for downstream datasets, including Flickr30k and COCO for image-text retrieval, VQAv2 and OKVQA for visual question answering (and ablation studies), and NLVR2 for visual reasoning.
Furthermore, we concentrated on optimizing neural network training and efficient tuning of multiple modalities. We focused on the landscape design of the logistic function and derived a novel sequence of strictly convex functions that are at least as strict as logistic loss. Our empirical analysis applies the proposed rooted logistic objective to multiple deep models across various classification benchmarks. These are conducted on the following four public datasets from the UCI machine learning repository: Wine, Ionosphere, Madelon, and Specheart. Our results illustrate that training with the rooted loss function converges faster and yields performance improvements. Furthermore, we demonstrate the applications of our novel rooted loss function in post-training quantization and generative modeling-based downstream applications, such as fine-tuning the StyleGAN model with the rooted loss. In particular, we trained on public datasets CIFAR-10/100 and fine-tuned on two public datasets including Tiny-ImageNet and Food-101. We use two public datasets for our image generation experiments with StyleGAN, including FFHQ and the Stanford Dogs. For the quantization of language generation tasks, we use three public datasets: WikiText2, Penn Treebank, and C4.
Lastly, we bridge vision foundation models and vision-language models on open-vocabulary segmentation. We propose a diffusion-based layer to combine features from different samples for segmentation. Our layer can be used to incorporate information from multiple modalities for segmentation purposes within a single pipeline with no recurrences or recursions in them. Our pipeline uses diffused vision foundation models and CLIP to inform features across samples for novel concept segmentation, such as in training-free cases. We then incorporate language modality within our framework using CLIP embeddings for cut guidance to enhance open-vocabulary semantic segmentation. Our empirical results show significant improvements on various public benchmark datasets, including Pascal VOC, Pascal Context, MS-COCO, ADE20K, and Cityscapes.
Future research directions are summarized in three key directions: efficient tuning of multimodal models, advancing the reasoning abilities of multimodal models, and applying VLMs to scientific data from different domains
Search for Pair Production of Vector Like Tops, Each Decaying to a Top and a New Scalar
No full textThis thesis attempted to answer a number of interesting questions in the search for new physics within the realm beyond Standard Model theory.
An innovative approach was implemented to model the expected small background in the signal region as a variation on the traditional ABCD strategy, motivated by the expectation of the shape of background photons in the signal region as some function of the shape of untagged photon objects in the control region. Ambiguity arising from the choice of the reconstructed top to be matched with the reconstructed new scalar boson in the process of reconstructing the vector like top demanded an innovative answer in order to provide a better solution than the coin-toss solution of the choice between leading and sub-leading jet.
This analysis provides a view into the signal and control region shapes and characteristics of Data and Standard Model MC backgrounds relevant to pair production of vector like quarks and is a first search for such pair production where the vector-like tops consequently decay to a new scalar boson and a standard model top
Effective and Collaborative Practices at HBCUs to Support Black Students with Disabilities
No full textThe purpose of this interview study was to explore the practices and collaborative efforts of Directors of Disability Services (DDS) at four-year Historically Black Colleges and Universities (HBCUs) used to address the needs of Black students with disabilities (SwDs). With more Black SwDs attending post-secondary education (PSE), understanding ways to better support them through the recruitment, retention, and completion phases is essential. In the study, 10 DDS participated in structured interviews to discuss what supports and organizations they have used or witnessed to promote educational access for Black SwDs and what barriers they have observed that hinder Black SwDs PSE success. DDS discussed strategies they used to ensure Black SwDs are successful in attending their HBCU, and some approaches included being proactive and reaching Black SwDs while they were in high school, partnering with HBCU support services on campus, and educating faculty, administration, and students on disability awareness. In addition, DDS described barriers that thwarted the progress of Black SwDs PSE outcomes. For example, from the findings, trauma, and stigma associated with disability labels, lack of funding and resources, and Black SwDs being underprepared for PSE hindered these results. Key findings included (a) being proactive and engaging Black SwDs early, (b) building relationships and creating inclusive campus environments, (c) DS visibility and the allocation of resources and funding, and (d) partnering with local, state, and national organizations to assist with supporting Black SwDs while they attend PSE. Further examination of limitations and implications for research and practice are presented
Optimizing Ligament Modeling Techniques in Total Knee Replacement Finite Element Models
No full textA total knee replacement (TKR) is an orthopedic implant designed to alleviate pain and restore joint function as an end-stage treatment for osteoarthritis. As the number of TKR surgeries performed continues to increase, it is important to understand the causes of TKR failure to improve patient outcomes and prevent TKR failure. TKRs failure modes include instability and malalignment which are influenced by ligament balancing. Poor ligament balancing can lead to instability and irregular kinematics, which can result in implant failure. Computational TKR research, including finite element analysis (FEA) modeling, is an efficient and low-risk technique for investigating knee biomechanics. Ligaments should be incorporated within TKR FEA models to investigate the correlation between subject-specific ligament anatomy and mechanical properties , ligament balancing, and TKR function. However, there is currently no consensus on the appropriate computational modeling choices to best represent these structures. Therefore, this thesis seeks to 1) quantify how knee ligament modeling choices affect resulting predictions of TKR kinematics and laxity, and 2) explore how knee ligament mechanical properties influence TKR function within a TKR FEA model. A sensitivity study of ligament attachment site found 14 of the 18 parameters (superior and inferior attachment point of each ligament, along 3 translational directions) were statistically significant for at least one kinematic motion. Knee laxity testing results demonstrated that a linear ligament material model resulted in smaller laxity ranges compared to the nonlinear ligament material model but followed a similar trend. A sensitivity study of ligament material parameters found 5 of the 6 parameters (stiffness and slack length of each ligament) achieved statistical significance for at least one kinematic motion. Lastly, increasing the number of ligament fibers from 1 to 5 influenced knee laxity and ligament engagement. These studies demonstrate that varying knee ligament anatomical and mechanical properties impact TKR kinematics and laxity. Study limitations include the use of one-dimensional representations of ligaments, one implant type, and the lack of direct comparison with cadaveric testing. This work provides a crucial step towards the development of patient-specific models to better predict TKR failure and improve surgical guidance
Evaluation of Sensor-Based Air Monitoring Networks in the U.S. and Globally: Guidance for Urban Networks
No full textAdvancements in air sensor technology have paved the path for community-driven environmental monitoring, particularly in areas underserved by traditional air monitoring networks. This study examines national and global air sensor networks, focusing on technological, administrative, financial, and community engagement aspects. Through expert interviews and case studies, it explores the best practices and lessons for design and implementation networks to guide similar efforts in urban industrial corridors with environmental justice (EJ) communities. A mixed-methods approach was applied to this research. A comprehensive review of scientific and grey literature was performed. Semi-structured interviews were conducted with air monitoring experts from diverse sectors, including academia, government agencies, private industries, and non-profit organizations. A rating scale evaluated the comprehensiveness of answers. Qualitative methods of interview transcripts identified patterns and emerging themes across several network elements regarding network design, stakeholder engagement, lessons learned, strengths, weaknesses, and factors to successful implementation. Successful air sensor networks require a thorough assessment of the existing sensor technology and monitoring gaps, structured operational and maintenance plans, a variable funding model, dynamic network design, cross-sector collaboration, broad stakeholder buy-in, and meaningful community engagement. Common challenges include sensor maintenance, data transmission, and calibration. Experts recommend structured management plans, variable funding mechanisms, and community programs for transparency, education, and data-driven action. Findings from this study can inform the development of a citywide sensor-based air monitoring network in Chicago, specifically in environmental justice communities
Helping Older Adults with Ongoing Mobile Technology Use: Challenges and Design Opportunities
No full textDespite increasing mobile technology adoption among older adults, challenges hinder their effective and continued application use. This dissertation explores the factors influencing older adults' ongoing mobile technology use and investigates ways to enhance their mobile technology experiences.
To better explore these challenges during older adults ongoing mobile technology use, we introduce the Senior Technology Learning Preferences Model for Mobile Technologies (STELE), which identifies two primary pathways for tech engagement: self-reliant support, where older adults learn independently through exploration and resources such as online tutorials, and social support, where they seek help from knowledgeable others. STELE explains that an older adult’s choice between these pathways is shaped by their identity as a tech learner, their past learning experiences, and their access to a supportive social network. Expanding on this, we analyze the relational dynamics of tech caregiving across three cultural contexts, identifying key challenges such as repeated requests, emotional labor, and the strategies learners and caregivers use to maintain positive interactions.
To further understand older adults' support preferences, we develop and validate the Mobile Tech Support Questionnaire (MTSQ), which measures older adults’ preference for and perceived quality of support during continued mobile technology use. Our findings reveal that preferences for self-reliant and social support are not mutually exclusive; older adults often rely on both, depending on the situation. The availability and reliability of support influence how frequently older adults seek that type of support and how easy they find the experience. We also investigate how feelings of confidence and anxiety during mobile use, as well as mobile device proficiency, play a significant role in shaping support preferences and perceptions. Findings suggest that higher proficiency and confidence predict a preference for self-reliant tech support, while proficiency also influences perceptions of its quality. These findings highlight the importance of considering individual learning styles and personalization when designing tech support tools for older adults.
Since effective tech support of any kind begins with clear problem communication, we investigate how older adults formulate their technology-related queries in unstructured contexts. A diary study reveals key communication challenges, including verbosity, incompleteness, over-specification, and under-specification. Using a few-shot prompt-chaining approach with GPT-4o, we evaluate how reformulating queries improves solution accuracy. Results show that even minor changes to the structure of a query significantly improve the accuracy of solutions generated by AI and retrieved by search engines. Evaluations by older adults and tech helpers further highlight the potential of this approach in real-life scenarios.
By integrating empirical research with practical design considerations, this dissertation provides a deeper understanding of how older adults learn and use mobile technology. We lay the foundation for developing more inclusive, socially aware, and effective support systems that empower older adults to navigate mobile technologies with confidence
Data sources:
Access Repository Dashboard
Do you manage Open Research Online? Become a CORE Member to access insider analytics, issue reports and manage access to outputs from your repository in the CORE Repository Dashboard! 👇