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
SWIFTopic: On-Device AI for Topic Modeling
No full textDespite advancements in topic modeling and embedding techniques, deploying these methods entirely on smartphones remains challenging. Current desktop frameworks like BERTopic rely on computationally intensive transformer-based models, making them unsuitable for devices with limited processing power. This limitation restricts the feasibility of high-accuracy, on-device topic modeling and raises privacy concerns, as data must be sent to remote servers for processing.
Although model compression techniques have introduced Mini Language Models by reducing transformer size and computational demands, their application in mobile topic modeling is largely unexplored.
This thesis presents SWIFTopic, a Swift-based framework designed to bring topic modeling inference onto iOS devices. SWIFTopic is inspired by BERTopic, a Python-based framework, and adapts its core inference tasks—such as tokenization, embedding, dimensionality reduction, and Euclidean distance calculation—for on-device execution. While training tasks, including clustering, weighting, and representation tuning, are performed on a desktop environment, the post-training artifacts (e.g., cluster centers, generated topics, and dimensionality reduction data) are then integrated into SWIFTopic for iOS inference.
This framework is applied to a healthcare use case focused on detecting suicidal ideation. A BERTopic model is first trained on a desktop using a dataset of suicide-related and non-suicide-related texts. The training outputs are then transferred to SWIFTopic for topic inference on iOS. Through prompt-engineered labels, the application classifies the inferred topics as either “suicide” or “non-suicide,” achieving a classification accuracy of 82%.
The dataset used is anonymized and sourced from publicly accessible platforms. It incorporates English Historical Quotes, Book Corpus, Simple Wikipedia LM, and posts from Reddit. The Reddit data specifically comes from Kaggle and other published research datasets, while the remaining data is sourced from Hugging Face. Reddit users are made aware that their posts are publicly accessible through Reddit’s Terms and Conditions. No personal identifying information (eg, names, locations, or IP addresses) were collected. Further, none of the authors participated in any discussions; thus, it was not necessary to inform users that their posts may be used for research. Because the collected data set is publicly available and already deidentified, it qualifies as “Not Human Research.”
Performance evaluation shows that SWIFTopic maintains a low memory footprint (0.9% RAM usage) and competitive latency (~500 ms) during inference. Although it is 14.5 times slower than remote inference, it trades off speed for privacy, providing acceptable response times while ensuring data security through local processing.
This thesis represents a significant advancement in on-device natural language processing and mobile software engineering, introducing a new framework that enables developers to build smarter iOS apps without compromising user privacy
Towards a Theory of Complex Mathematical Personhood: Math Teachers of Color in the Workplace
No full textThis qualitative study examines the experiences of veteran mathematics teachers of color who choose to remain in the profession despite systemic racial and structural barriers. Veteran is defined as a teacher who has been in the profession for 5 or more years. Motivated by both personal and professional experiences of racial injustice, the research seeks to understand how teaching within White institutional spaces and how teachers of color navigate and construct their teaching identities. The study proposes and constructs a theory of complex mathematical personhood, where VMToCs evolve these identities over time. The study is informed by an intersectional, critical race feminist framework and employs portraiture methods (Lightfoot & Hoffman-Davis, 1997) to illuminate the macrostructures of oppression in mathematics education.
The study presents the lived experiences of seven veteran mathematics teachers of color, who collectively bring a century of teaching experience. Their portraits reveal how they leverage their racialized and gendered experiences to foster critically conscious approaches to math education. Through these portraits, the study sheds light on how teachers of color engage in acts of resistance and reimagine what it means to teach mathematics in ways that challenge dominant narratives
Demographic Characteristics and Multiple Mini Interview Performance of Applicants to Medical School
No full textThis study evaluates associations between sociodemographic characteristics of medical school applicants and performance on Multiple Mini Interviews (MMI). Applicants who participated in MMI as part of their application to the University of Illinois College of Medicine had their scores collected and averaged. The scores were recorded on an electronic platform which also contained sociodemographic information. Descriptive statistics were used to evaluate data trends for each subgroup and across subgroups of subjects. Differences in MMI scores based on age, sociodemographic status, and gender were compared by each racial and ethnic group. Analysis of variance was used to compare the MMI scores across the various groups. Multiple linear regression was used to analyze how independent demographic variables affected the MMI score. The cohort of subjects in this study was large and diverse, which allowed for a detailed analysis of the effects sociodemographic characteristics have on MMI performance.
Using data from five years (admissions cycles from 2020-2024, n = 3,447), this study shows that male gender and disadvantaged statuses are associated with lower MMI performance, and there is no difference in MMI performance based on age. When evaluating MMI performance based on racial and ethnic status, Asian and Black or African American applicants scored higher on MMI, while white applicants performed lower. There was no difference in performance based on ethnicity. For Asian applicants, male and disadvantaged statuses were associated with lower MMI performance. For Black or African American applicants, younger age and male gender were associated with lower performance, but disadvantaged status was not. For Native American and Alaskan Native applicants, younger age was associated with higher MMI performance, while lower MMI performance was associated with male Native Hawaiian or Pacific Islander applicants. There was an association with lower MMI scores for white applicants who were male or disadvantaged. Hispanic or Latino subjects showed an association with lower MMI performance for those who identify as male.
In conclusion, this study shows that there are differences in MMI performance among various sociodemographic groups. These results provide insight that can inform the use of MMI as an evaluation tool for medical school admissions
Induced Model Matching: Learning from Restricted Models
No full textGiven a very good predictive model that uses a restricted feature set, what is the best way to incorporate it into a large full-featured model? There are two main scenarios where this problem can arise: (a) we have ample amounts of data with restricted features or (b) the restricted feature model is easier to learn with existing data. Here are some relevant cases. For (a), while training a logistic regression model most data may be missing many features for privacy reasons or while training an MDP policy with expensive data (with full sensing), we may have a past model built using a lot of cheap data (partial sensing) available. For (b), it is common to augment LM data with smaller language models, such as N-grams, and these are assumed to be reliably buildable from the same data. We discuss prior works that have used restricted models in the training of full-featured models using implicit or explicit regularization and we reveal their caveats. To solve these caveats, we propose our methodology, Induced Model Matching (IMM), that aligns the context-restricted, or induced, version of the large model with the restricted model. We show that correctly incorporating the restriction is crucial to have consistency in the limit (theoretically) and to achieve better performance with finite samples (experimentally) than the past approaches. Namely, these past approaches are (1) noising, which is implicit in addressing the problem, and (2) reverse knowledge distillation from weak teachers, which is explicit. These past approaches do not exploit the restriction being the nature of the weakness and can be problematic in terms of consistency. We demonstrate the merits of IMM using logistic regression as a proof of concept. We then apply it in language modeling (the application that initially inspired it) and demonstrate it on both LSTM and transformer full models, using bigrams as restricted models. We lastly give a simple RL example, which shows that POMDP policies can help learn better MDP policies. The IMM principle is thus generally applicable in common scenarios where restricted data is cheaper to collect or restricted models are easier to learn
The Role of Orthodontists in the Continuum of Craniofacial Microsomia Care: Survey of Craniofacial Teams
No full textObjectives: Craniofacial microsomia is a developmental defect that presents as an asymmetric development of craniofacial structures derived from first and second branchial arches. There is a wide variety of clinical presentations and severities to the condition. There is great variation in treatment protocols and timing among clinicians, particularly in the preference of surgical intervention before or after skeletal maturity. This study aims to identify craniofacial team related factors that are associated with the treatment protocols that they use and identify commonalities in various treatment protocols and establish a standardized treatment approach that can be used to treat patients with CFM. From the treatment protocols identified, we aim to qualitatively delineate the role of orthodontist in the continuum of craniofacial microsomia care and the various orthodontic and surgical interventions and their timing followed by the craniofacial teams across the United States.
Methods: A 15-item questionnaire was distributed electronically to 121 craniofacial teams across the United States via Qualtrics Survey Software. The survey included 3 major blocks: Treatment Characteristics, Diagnosis, and Treatment of CFM. The questions explored craniofacial center characteristics, case load, diagnostic tools and systems, as well as preferences in surgical and orthodontic treatment methods. 22 respondents completed the full survey, thus resulting in an 18% response rate. Descriptive statistics were obtained to summarize the data Chi Square tests and Fisher’s Exact test were used to find any associations between predictor and outcome variables. All statistical tests were two-sided and statistically significant at a p-value of <0.05.
Results: A statistically significant association was found between status of academic accreditation and number of active orthodontic cases, where non-academically accredited centers had fewer active orthodontic cases. No other statistically significant associations were obtained.
Conclusions: Most respondents reported performing mandibular surgery after the pubertal growth spurt or at skeletal maturity. However, several participants reported an exception where early surgery may be considered during growth to control or improve facial asymmetry if the patient is experiencing significant psychosocial challenges compromising their quality of life. This sheds valuable light on the potential indications of early versus late surgical intervention. Most respondents reported using traditional orthodontic brackets with functional appliances and mini-implants as adjunctive treatment modalities to achieve orthopedic and orthodontic treatment objectives. The results of this study provide a framework for future studies to define a clear standard of care for craniofacial microsomia cases
Assessing the Constituency-Based Pledges in Pledge Making and Fulfillment in the United States and Turkey
No full textThis dissertation explores pledges made by the two main parties in the United States between 2001 and 2016 and the two main parties in Turkey between 2002 and 2015 to understand the role of constituency-based pledges and pledges made to various constituency groups in pledge making and pledge fulfillment in these societies. By utilizing descriptive analyses and relying on theories of partisan sorting and partisan polarization concerning parties’ pledge-making patterns, findings demonstrate that political parties make a considerable number of pledges targeting a specific constituency group. By using descriptive and predictive analyses and building on theories of punctuated equilibrium and voters’ perception of campaign promises (promise keeping and promise breaking) with respect to parties’ pledge fulfillment patterns, findings show that the likelihood of pledge fulfillment is lower for constituency-specific pledges than for mass appeal pledges in the US context and the constituency variable is not statistically significant in the Turkish case. These results pose grave concerns about the representation dynamics of certain groups and are indicative of potential disadvantages of high and asymmetric pledge fulfillment of certain groups at the expense of others, thereby potentially aggravating group-level inequalities. Accordingly, whereas high pledge fulfillment would be desirable from a democratic responsiveness perspective in this chain of program-to-policy linkage, it might potentially exert negative effects on certain groups that could not reap as many benefits as others
COVID-19 Clinical Outcomes in Hospitalized Patients with Obstructive Sleep Apnea
No full textIntroduction: COVID-19 is caused by the novel coronavirus SARS-CoV-2. Existing literature indicates that obstructive sleep apnea (OSA) is an independent risk factor for poor COVID-19 outcomes, and a pro-inflammatory state and immune dysregulation have been suggested as mechanisms underlying both conditions.
Objective: The aim of this retrospective matched case-control study is to evaluate clinical factors associated with COVID-19 severity in hospitalized patients with a clinically documented history of OSA compared to a control group of hospitalized COVID-19 patients with no documented history of OSA.
Materials and Methods: The UIC COVID-19 Registry for Research (Registry), a database containing demographic and a breadth of longitudinal clinical information of the first case series of patients who were hospitalized at UI Health between January 1 and May 6, 2020 and tested positive for COVID-19 between March 20 and April 30, 2020, was examined to identify patients with a documented history of OSA. Patients with OSA were matched 1:1 to control patients from the same Registry with no documented history of OSA. Cases and controls were matched based on: sex, age ± 3 years, race, ethnicity, and BMI category. Outcomes included symptoms at diagnosis, oxygen support, ICU admission, vitals, blood oxygen saturation, inflammatory biomarkers, inpatient length of stay, readmission beyond 20 days after testing positive until August 30, 2020, mortality, and CPAP use and adherence were selected to represent COVID-19 severity and were collected from the Registry and from electronic medical records.
Results: Of the 251 patients in the Registry who were hospitalized at UI Health, 41 cases with a documented history of OSA and 41 matched controls with no documented history of OSA were identified. Based upon matching, there were no statistically significant differences in demographic variables between cases and controls; overall, mean age was 57.5 ± 12.0 years, 48% were male, 63.4% were Black, 69.5% were non-Hispanic/Latino, and 75.6% had a BMI category of obese. Patients without OSA versus with OSA, respectively, were more likely to present at baseline with respiratory distress (41.5% vs. 19.5%; p=0.03) and chills (43.9% vs. 22.0%; p=0.03). Patients with OSA versus without OSA, respectively, were more likely to require oxygen support via nasal cannula (65.9% vs. 43.9%; p=0.046) and non-invasive ventilation (9.8% vs. 0%; p=0.04) as the maximum O2 support needed during admission. No statistically significant differences were found between inpatient length of stay, ICU admission, baseline levels of inflammatory markers, number of readmissions, and mortality.
Conclusion: Despite being less likely to present with respiratory distress, patients with OSA were more likely to require advanced levels of oxygen support during admission. Future research will be required to evaluate the role of potential respiratory compensations in patients with OSA, particularly in regard to respiratory challenges such as COVID-19
Peripheral Pattern Electroretinography: System Upgrade, Novel Protocol Development, & Clinical Validation
No full textClinical motivation: Glaucoma is the second leading cause of blindness worldwide affecting more than 60 million people. It almost always begins silently and asymptomatically causes irreversible damage and apoptosis to the retinal ganglion cells (RGC). Unfortunately, there is no cure for glaucoma but with therapeutic interventions involving monitoring, medications, and/or surgery it is possible to halt and significantly slow down the progressive loss of vision. Early detection and diagnosis are key and the most quintessential steps to preserving vision that have been or can potentially be affected by glaucoma. Capability gap: The standard of care diagnostic tests for glaucoma, OCT imaging and visual field tests, target the central retina. A less-used functional test, pattern electroretinography (PERG), also evaluates the central retina and has been shown to be very sensitive in detecting early signs of dysfunction. However, it has been hypothesized that the earliest retinal damage in glaucoma begins, at least in some patients, in the peripheral retina. Solution: To address this, an initial peripheral-retina PERG (pPERG) system prototype was previously built. For earliest detection, it is crucial to be able to detect glaucomatous damage in localized regions of the retina, which is time consuming and therefore clinically impractical if done serially. To address this, this study rebuilt the initial pPERG prototype and created a novel protocol for evaluating three radial sectors of the peripheral retina simultaneously. The system was validated on 26 subjects including glaucoma suspects, early-stage patients and normally sighted individuals. In summary, the novel pPERG combined with the central PERG have the potential to be a comprehensive diagnostic tool for glaucoma
Repressors’ True Emotions: Identifying Repressive Coping Style via an Implicit Measure of Emotions
No full textRepressive copers are defined as people who tend to avoid negative/ threatening information automatically (Myers, 2010). Via this process, repressive copers are not consciously aware of their own anxiety (Weinberger et al., 1979). This leads to large-scale implications on all self-report emotion scales’ accuracy. This project was aimed at replicating the foundational study on repressive copers, Weinberger et al (1979), to support the need to account for this population in research. Additionally, since explicit emotional measures are inaccurate for repressive copers this study tested an alternative implicit method of measuring emotions accuracy following a threat manipulation. Weinberger et al (1979) determined that repressive copers were significantly slower when responding to threatening prompts when compared to the other groups, high and low anxious. To replicate this finding, participants were asked to complete a series of neutral and threatening prompts verbally, while under the impression they were being recorded though multiple methods. Following both sets of prompts participants were asked to complete the Implicit Measures of Discrete Emotional States (IMDES). The results of this study failed to replicate Weinberger et al (1979) as there were no group differences in response time difference scores found. There was a main effect of prompt from neutral to threatening prompts. As for the IMDES, group differences were seen in sadness and happiness, which was not in line with the hypothesis. The expected difference in anxiety reporting from neutral to threatening trials was not observed. One limitation to the study was that the trial prompt was not targeted at one clear emotional state. This left the possibility for other emotions to be more salient than expected. The results of this study make it clear that more work needs to be done evaluating previous findings on repressive copers
Computational Modeling of Rail-Induced Vibrations: A Predictive Framework for Building Response
No full textRailways play a crucial role in transportation, facilitating the efficient movement of goods and passengers over long distances. However, the vibrations generated by passing trains can pose significant challenges to nearby residential communities and buildings. These vibrations primarily lead to occupant discomfort, disrupting daily life and potentially impacting health and well-being. Over time, prolonged exposure to such disturbances may contribute to structural fatigue, making developing predictive models that assess and mitigate these effects essential.
A predictive modeling approach has been developed to estimate railway-induced vibrations and their impact on surrounding structures to address this issue. This model operates under the principles of modal superposition, where a modal analysis is performed to determine the system’s dynamic behavior. The rail motion is then solved using multibody dynamics, effectively coupling finite element analysis with multibody system simulations. This approach offers significant advantages over traditional computational techniques by enhancing computational efficiency while maintaining accuracy in capturing the complex interactions between train dynamics and structural response.
Such a model could be instrumental for evaluating and mitigating railway-induced vibrations. It could provide insights into how tracks’ modifications influence vibrational impacts, helping to develop effective mitigation strategies. Additionally, this model could serve as a foundation for optimizing infrastructure layouts & improving building designs near railway corridor
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! 👇