10:20 AM – 11:35 AM | Merten Hall, Rooms 1202, 1203, 3300
AI and Society: Case Studies
10:20 AM – 11:35 AM | Merten Hall, Room 1202
Discussant: Dr. Anthony Kelly
Integrated Host and Pathogen Urinary Proteomics Reveal Symptom-Linked Biomarkers of Borreliosis
Sindhu Datla (College of Science)
Background: Lyme disease remains difficult to diagnose due to heterogeneous clinical presentations, delayed serologic conversion, and the absence of reliable direct biomarkers. There is an essential demand for non-invasive, direct diagnostics that accurately reflects active infection and symptom burden. Methods: A longitudinal cohort of 66 patients with suspected tick-borne illness was monitored for up to one year. Symptom burden was assessed at each visit using the General Symptom Questionnaire-30 (GSQ-30). Standard serologic testing for Lyme disease and B. miyamotoi was performed. Urine samples were subjected to mass spectrometry-based proteomic analysis to detect Borrelia-specific peptides. Multivariable linear regression models were used to evaluate associations between peptide abundance, clinical variables, and symptom severity. Human host proteomic profiles were analyzed using LASSO regression to identify disease-associated protein signatures. A user-centered diagnostic urine-collection device was also developed and evaluated through a QR-based usability survey. Results: In the study group, 38% were Lyme-seropositive and 15% were B. miyamotoi–positive, with no overlap between groups. A total of 105 Borrelia-derived urinary peptides were identified. After adjustment for age, sex, and serologic status, Borrelia peptide abundance showed a strong positive association with GSQ-30 symptom scores (β = 0.57, p < 0.00001), indicating that increased symptom burden correlates with higher microbial peptide detection. Computational analysis of human proteomic data using LASSO regression identified candidate host proteins associated with disease-related biological pathways. Conclusions: Urinary mass spectrometry detection of Borrelia peptides provides a sensitive, noninvasive correlate of symptom burden in Lyme disease and outperforms serology alone.The ongoing development of a portable urine collection device aims to translate these data into practical, patient-friendly diagnostic platform.
“We really don’t know all the effects”: Unmasking Environmental In(justices) in digital infrastructures in Northern Virginia (NoVA) through infrastructural storytelling
Isaac Newton (College of Humanities and Social Sciences)
The field of technical and professional communication has long contributed to technology and environmental policymaking. Extant studies call for sustainable practices of activism and social justice to expose the hidden expedient rhetorics that have been socially privileged in technical communication. This proposal stems from the unfettered expansion of data centers in NoVA despite residents’ resistance (Lifset et al., 2025). NoVA is considered a hub for data centers, with over 70% of the world’s internet traffic passing through its infrastructure. The current trajectory of Artificial Intelligence (AI) development perpetuates environmental injustices, with little regard for rebound effects on the environment and digital ecosystems. Maintaining the efficiency of AI technologies requires building data centers to train and deploy AI models. This perspective reflects a utopian view of AI that “Data centers will grow so efficient, their impact will stop being a problem; generative AI will unlock new climate innovation; AGI will solve climate change once and for all” (Hao, 2025, p. 276; Crawford, 2021). Such efficiency rhetoric undermines environmental sustainability efforts and deepens the vulnerability of local communities. Using infrastructural storytelling (Edwards, Gelms, & Shivener, 2023), the study investigates the environmental injustices of data center development in NoVA through “the lived realities of the people and ecosystems where digital infrastructures reside” (p. 243). The aim is to deconstruct the power dynamics in data center development and disrupt the corporate narratives cloaked in efficiency rhetoric. Specifically, the study focuses on community narratives about the environmental impacts of data centers in NoVA, as reported in news stories and on social media platforms.
Natural Language Processing for Extreme Environments: Comparative Analysis of Human Behavior Literature
Abdullah Almalki (College of Science), Dr. Anamaria Berea
This paper presents a comparative analysis of human behavior in extreme environments using Natural Language Processing (NLP) techniques. We analyze a corpus of 78 research papers on human behavior in extreme environments, comparing computational NLP analysis with human expert annotations. Using transformer-based embeddings for topic modeling, we identify five key behavioral domains: team dynamics, environmental factors, polar environments, occupational health, and social networks. Our comparative framework reveals both alignments and divergences between computational and human expert analyses, demonstrating the value of NLP for extracting structured insights from qualitative research literature. This work demonstrates how NLP can bridge qualitative research and computational analysis, providing a methodological framework for translating domain literature into structured knowledge. The findings have implications for understanding human behavior in extreme environments and developing more effective support systems for teams operating in such conditions.
Evaluating AI Assistants Usability for Enhancing Student Learning in Machine Learning Classification
Sumaya Binte Zilani Choya (College of Engineering and Computing), Mihai Boicu
As artificial intelligence assistants (AIAs) become increasingly embedded in data science and machine learning (ML) education, evaluation has largely focused on task completion and output correctness rather than their effectiveness as instructional tools. This study addresses that gap by systematically examining how popular AIAs support student learning across the full machine learning workflow, with particular emphasis on conceptual understanding, analytical reasoning, and explainability. Using twenty standardized sentiment-classification prompts grounded in core ML principles, six widely used AIAs, ChatGPT, Gemini, Claude, DeepSeek, Grok, and Copilot, were evaluated across four datasets: IMDB Movie Reviews, Twitter Airline Sentiment, Amazon Fine Food Reviews, and Financial News Headlines. Each response was independently assessed by three researchers using a structured rubric measuring theoretical understanding, explanation quality, practical implementation, depth of analytical reasoning, overall understanding, educational communication style, and explainability verification. Across four sentiment-analysis datasets and 80 standardized prompts, AIAs achieved consistently high scores in implementation with practical integration (mean ≈ 21–23 out of 25), while explainability verification emerged as the weakest dimension, with mean scores ranging from 2.58 to 4.00 out of 5 across systems. Notably, Claude (4.13 out 5) and ChatGPT (4.12 out of 5) attained the highest overall understanding scores, demonstrating stronger integration of reasoning, explanation, and evaluation compared to other assistants, whose performance varied substantially by dataset and pedagogical criterion. However, it maintained solid performance in implementation-oriented tasks. These results support prompt-based AI structuring for deeper conceptual learning and demonstrate task-dependent pedagogical diversity across AIAs. Thus, the selection of an artificial intelligence assistant (AIA) will depend on the specific educational objectives and the dimension to emphasize by its use.
Environmental Futures: Interdisciplinary Pathways
10:20AM – 11:35AM | Merten Hall, Room 1203
Discussant: Dr. Lucas RF Henneman
Reevaluating SST Gradient Trends Using ARX-Derived Internal Variability
Aahelee Sarker (College of Science), Dr. Timothy DelSole
In this study, we introduce a novel approach for separating forced trends, seasonal cycles, and internal variability in climate time series. The method is based on autoregressive models augmented with external forcing terms to represent the annual cycle and climate change signals, resulting in an ARX model-an autoregressive model with exogenous input. Applying this method to sea surface temperature (SST) data from Coupled Model Intercomparison Project Phase 6 (CMIP6) historical simulations (1958–2014), we find that applying an ARX model to a single ensemble member can effectively capture the externally forced signal traditionally derived from ensemble means, with the ARX model’s forced response and noise variance closely aligning with ensemble-based estimates. This suggests that the ARX model provides a reliable framework for characterizing the internal dynamics of climate variables, based solely on single realization. We apply this model to diagnose the discrepancy between CMIP6 projections of a future weakening in the west-to-east zonal sea surface temperature (SST) gradient and the observed strengthening since the mid-20th century. When applied to observations, the SST gradient trend falls within the range of internal variability, indicating that the observed trend may result from natural variability in the region. Surprisingly, our result also suggests that the observed trend is not inconsistent with model projections once the range of internal variability is accounted for.
Solving Latency Location-Routing Problem using Dynamic Programming method under uncertainties
Peng Ren (College of Engineering and Computing), Dr. Rajesh Ganesan, Department of Systems Engineering and Operations Research
My research of Latency Location-Routing Problem (LLRP) addresses the joint optimization of depot selection and vehicle routing with the objective of minimizing cumulative customer waiting time (latency). This problem is particularly critical in disaster response and humanitarian logistics, where timely service under evolving operational constraints directly affects recovery outcomes. However, much of the existing literature relies on static assumptions, often neglecting time-varying demand, stochastic system evolution, and the need for adaptive decision-making in dynamic environments. This research develops a dynamic programming (DP) formulation of the LLRP that explicitly captures multi-stage system evolution under stochastic demand. The system state is defined by depot configurations, vehicle locations, remaining vehicle capacities, and time-dependent customer demand. A Bellman recursion is used to optimize sequential location and routing decisions with the objective of minimizing total accumulated latency over time. As for numerical results, one small-scale case study demonstrates how the proposed formulation accommodates newly arriving demand and updates vehicle states across decision stages, enabling interpretable and adaptive routing behavior. While exact DP becomes computationally intractable for large instances, the formulation serves as a principled foundation for improving solution quality and computational efficiency through approximate dynamic programming. In particular, this work focuses on enhancing reinforcement learning and value-iteration–based methods to better approximate the value function and accelerate convergence. The proposed framework is currently being extended to flood-disaster scenarios, with the goal of developing adaptive disaster-relief routing strategies that support effective response and post-flood recovery under uncertainty.
Media content analysis of social media posts on the X platform highlighting community activism in the wake of a climate-induced disaster in Gilgit Baltistan, Pakistan
Syed Muhammad Abubakar (College of Humanities and Social Sciences)
We conducted a media content analysis of 150 social media posts on the X platform (formerly Twitter) regarding the community activism of three shepherds to timely alert communities regarding a Glacial Lake Outburst Flood (GLOF) event in Ghizer district of Gilgit Baltistan, Pakistan, on August 22, 2025, which saved around 300 lives. A sample of 150 social media posts on X was selected using a purposive sampling technique. We adopted a mixed-methods approach, whereby the Situational Crisis Communication Theory (SCCT) was used for quantitative research (RQ1 and RQ2), and an inductive approach was used for qualitative research (RQ3). RQ1 focused on whether the early warning messages sent by the shepherds to alert communities regarding the GLOF event received appreciation from the general public on X. RQ2 focused on whether the social media posts that featured Pakistan’s PM appreciating the shepherds for their heroism received more likes, shares, and comments, versus others. RQ3 focused on whether the posts highlighted the devastation caused by the GLOF and potential threats to communities. Our quantitative research findings confirm significant associations between variables using chi-square tests and t-tests with SCCT as the theoretical framework . Whereas the qualitative research findings highlight the devastation caused by the GLOF event, the worsening climate crisis, threats to communities, and criticism of the government and civil society on X. Future research can include other social media platforms and may also include other Hindu Kush Himalaya (HKH)-member countries in the dataset for broader perspectives from the Global South.
Discerning the Role of Latent Heat Release in Changes of Observed Extratropical Cyclones in the Atlantic and Pacific Storm Tracks
Austin Reed (College of Science), James L. KinterDiagnosing how global warming has impacted the role of Latent Heat Release (LHR) in Extratropical Cyclone (ETC) development remains unclear. While increased atmospheric moisture can intensify individual ETCs through LHR, the timing and magnitude of these effects may differ between the Atlantic and Pacific storm tracks. This work adopts the Multiple Object Tracking algorithm to identify Atlantic and Pacific cool-season ETCs from 1965-2023. ERA-5 storm-relative composites and an Ertel Potential Vorticity budget are used to quantify the influence and change of LHR for each basin. Atlantic ETCs show expansive significant increases in low-level moisture across the entire lifecycle, yet modest increases in LHR (0.01-0.02 PVU/12 hours) distributed both east and west of the storm center. In contrast, Pacific ETCs exhibit more spatially concentrated significant increases in low-level moisture near the storm center, particularly in the meridional direction 24 hours prior to peak intensity. This concentrated moistening drives nearly double the LHR change compared to Atlantic ETCs (0.02-0.04 PVU/12 hours), focused east of the center with peak signals 12-24 hours before maximum intensity. These contrasting behaviors appear linked to significant increases in vertical motion throughout the atmospheric column for Pacific ETCs that spatially correlate with moisture changes, a signal notably absent in Atlantic storms. Improved process understanding of LHR in ETCs can enhance prediction of damaging ETC impacts like extreme precipitation on weather and climate timescales.
Mental Health Across Disciplines: Identity, Society, and Care
10:20 AM – 11:35 AM | Merten Hall, Room 3300
Discussant: Dr. Shekila Melchior
The Prevalence of Poor Behavioral Health Among College Students by Gender, Sexual Orientation, and Racial Identity: The Role of Discrimination and Microaggressions
Tolu Okuneye (College of Public Health), Andrew Godley, Elaine C. Russell, Lisa L. Lindley, Kenneth W. Griffin
Background: College students face unique stressors that increase their risk for poor behavioral health. Experiences of discrimination and microaggressions on college campuses, and especially among students with minority identities, further worsen this risk. Methods: We assessed the association between experiences of discrimination and/or microaggressions and poor behavioral health (an index of severe psychological distress and substance use dependence risk) among college students in Spring 2023 (N=30,880) using cross-sectional data from the American College Health Association’s National College Health Assessment III (ACHA-NCHA III). Results: Poor behavioral health was reported by 42.9% of students, and the prevalence of discrimination and microaggressions in the sample was 11.3% and 18.6%, respectively. Students with minoritized identities who experienced discrimination or microaggressions were more likely to report poor behavioral health. In the logistic regression models, non-binary students (OR = 2.28, CI: 2.06, 2.54, p < 0.001), non-heterosexual (OR = 2.37, CI: 2.26, 2.49, p < 0.001), Whites (OR = 1.25, CI: 1.20, 1.31, p < 0.001) and Multi/Biracial (OR = 1.21, CI: 1.12, 1.30, p < 0.001) students were more likely to report poor behavioral health compared to the reference group, holding discrimination and microaggressions constant. We found that the interaction effect between race or ethnicity and experiences of discrimination and/or microaggressions was significant for White and Asian/Asian American students. Conclusion: While experiences of discrimination and microaggressions are more common among students with minoritized identities, these experiences can contribute to poor behavioral health among all youth, irrespective of minority group status. Efforts to increase resilience on college campuses may improve behavioral health.
The Effect of Women’s Economic Rights and Legislative Representation on National Happiness: A Global Analysis
Esha Doshi (Schar School of Policy and Government), Katherine M. Chiriboga, Jessica Terman, Nicole E. Proto
Although worldwide efforts to increase women’s presence in national legislatures have expanded dramatically, clarity regarding their impact on broader societal progress remains elusive. This study explores whether a positive correlation exists between a higher percentage of women in the lower house of legislature, greater women’s economic rights, and overall national happiness as measured by the World Happiness Report. Countries were grouped into quartiles and analyzed via a one-way ANOVA to compare mean national happiness. Controlling for the strength of women’s economic rights, a multivariate regression was also conducted. Results indicated no statistically significant difference in national happiness based solely on women’s representation in the lower house; however, a significant positive correlation emerged between women’s economic rights and reported happiness. Other contextual factors, such as quota systems arising from war or instability, may serve as confounding variables that influence the relationship between political representation and well-being. These findings challenge the assumption that gender parity alone can create societal outcomes that elevate national happiness above other nations. This research underscores the importance of scrutinizing the underlying political, structural, and economic conditions that shape societal outcomes. Whereas economic rights have a more immediate impact on well-being, legislative representation cannot shift national priorities in a vacuum. Ultimately, meaningful equity efforts and overall societal stability are more critical for the inclusive governance required to move the needle on national happiness.
Emotional Intelligence as a Leadership Competency in the Public Sector
Kunj Malhotra (Schar School of Policy and Government)
Emotional intelligence (EI) is increasingly recognized as a critical competency for public‑sector professionals who operate in environments shaped by high stress, public accountability, and complex service demands. This project examines how EI contributes to leadership effectiveness, employee well‑being, and citizen‑facing service delivery within U.S. government agencies. The purpose of this study is to understand how EI can address persistent challenges in public administration, including burnout, communication breakdowns, and declining public trust. The central research question guiding this project is: How does emotional intelligence influence leadership effectiveness and organizational outcomes in the U.S. public sector? This study uses a structured conceptual analysis grounded in interdisciplinary scholarship from psychology, public administration, and organizational behavior. It synthesizes foundational EI frameworks developed by Mayer and Salovey and Daniel Goleman, alongside contemporary research on public‑sector leadership and organizational culture. The method focuses on identifying patterns across existing studies rather than generating new empirical data. Findings from the literature indicate that EI is consistently associated with improved communication, stronger team cohesion, reduced burnout, and higher levels of trust in leadership. Research also suggests that EI‑based training programs can enhance conflict management and service responsiveness, though implementation across agencies remains uneven. The analysis concludes that EI is an underutilized but high‑impact competency in the U.S. public sector. Strengthening EI through leadership development, hiring practices, and organizational training may improve workplace climate and public trust. This project is ongoing and will expand to include qualitative interviews with public‑sector employees to explore practical barriers to EI adoption.