Graduate Division Summer Scholar Presentations

Faculty Mentor: Dr. Thomas Weimbs | MCDB
Graduate Mentor: Nick Holznecht
Postdoc Mentor: Melina Messing

Presentation Title:
Crystal Clear? Investigating Atypical Protein Kinase C Zeta’s Role in Renal Calcium Oxalate Crystal Clearance

Faculty Mentor: Dr. Joel Rothman| MCDB
Graduate Mentor: Samantha Fiallo

Presentation Title:
Investigating the Effects of Deleterious mitochondrial DNA on Phenotype and Cellular

Mitochondrial disease, usually caused by mutations in mitochondrial or nuclear DNA, affects various high-energy demanding systems in the body, including muscles and the nervous system. Because mitochondrial DNA (mtDNA) has limited repair mechanisms, it is more susceptible to mutations which can result in dysfunctional proteins and affect the production of cellular energy. This summer I am researching the effects of deleterious mitochondrial DNA (∆mtDNA) on health/fitness measured by brood size, egg laying rate, and induction of the unfolded protein response (UPRmt), a mitochondrial repair mechanism. The unfolded protein response (UPRmt) is a mitochondrial repair mechanism triggered by protein dysfunction and regulated by the transcription factor ATFS-1. This mechanism works to restore mitochondrial health on the protein level. However, in the presence of a uaDf5 mitochondrial mutation, when ATFS-1 is deleted, levels of the deleterious mitochondrial DNA (∆mtDNA) decrease significantly. Thus, UPRmt is observed to paradoxically maintain the mutation in a heteroplasmic state. Since UPRmt is triggered by protein dysfunction, my research aims to understand how different ∆mtDNA affect UPRmt induction and thus play a role in the mutation’s maintenance. Processes that regulate mtDNA quality control and purifying selection are most active during development into adulthood. Thus, conducting fitness tests on worms with ∆mtDNA of varying severity of protein dysfunction will help us better understand how it affects worm health and more information about how the mutation may be maintained in heteroplasmic cells which can help improve treatment for mitochondrial disease caused by mitochondrial mutations.

Faculty Mentor: Dr. Arnab Mukherjee | Chemical Engineering
Graduate Mentor: Jinyang Wan
Postdoc Mentor: Kaamini Dhanabalan

Presentation Title:
Genetically encoded Calcium Sensors for MRI applications

The visualization of cellular processes in vitro largely depends on genetically engineered fluorescent proteins. Green fluorescent protein (GFP) and mCherry commonly serve as biosensors that are sensitive to enzyme activity, pH changes, and various other cellular phenomena. Fluorescent proteins have provided valuable insights into cellular and molecular processes in living cells, however, there are inherent chemical limitations that hinder deeper physiological understanding. Their reliance on oxygen as a substrate for light emission restricts their effectiveness in hypoxic environments, rendering cellular activity in regions such as the gut microbiome, certain brain structures, and upper lymphoid tissues nearly “invisible”. Insufficient light penetration in thicker tissues limits fluorescence-based optical imaging, leaving many biological processes in these areas largely uncharacterized. To combat these shortcomings, our lab employs the Calcium Specific Protease Reporter (CASPR) system, which is a genetically engineered calcium sensing system for magnetic resonance imaging. We use aquaporin (Aqp1) as a biosensor to indicate calcium influxes, increasing the water diffusivity of the affected cells. We used split protease attached to calcium-binding moieties that are reconstituted in the presence of calcium, combining the split Tobacco Etch Virus (TEV). When the neuron is activated, Aqp1 is translated through the TEV protease. The increase of water diffusion creates a contrast on the MRI, after encoding background-subtraction to phase out non experimental cellular interactions. Future experiments will strive to define key signaling pathways that are currently poorly understood, and have potential to unveil the cellular mechanisms underlying various diseases and disorders.

Faculty Mentor: Dr. Brandon Woo | Psychological & Brain Sciences

Presentation Title:
How Does Exposure to Multiple Languages Relate to Perspective-Taking in Children?

When children are exposed to multiple languages, they may learn that the same language may not be understood by different people. This understanding of how others perceive things generalizes to mental state reasoning, or the ability to infer what others are thinking. Psychologists have found evidence that children with exposure to multiple languages may better represent whether or not others can see objects, overcoming differences in perspectives. Here, we ask whether exposure to multiple languages may also predict children’s understanding of others’ visual experiences: how a picture looks to someone else (e.g., whether a picture is upright or upside down). From a completed experiment, we have data on children’s verbal reasoning about others’ visual experiences and children’s language exposure. We found that children performed better in groups that heard 10-40% English, 60-90% English, and 90-99% English. 3 -year-olds who heard 90% English or more at home did worse on the perspective-taking task than children who heard English 90% of the time or less. Overall, 4-year-olds performed better than 3-year-olds.

Faculty Mentor: Dr. Nick Nidzieko | Geography
Graduate Mentor: Luke Carberry

Presentation Title:
Understanding Physical Properties within Kelp Forests

Kelp forests are critical ecosystems that provide habitat for keystone species and support the biodiversity of coastal regions. These productive ecosystems are influenced by the surrounding physical environment, including the transport of nutrients through different flow dynamics. Physical properties of the water column, like temperature and salinity, act as tracers to identify how masses of water have traveled vertically and horizontally in the water column. By observing the relationship between flow dynamics in and around kelp forests, connections can be made between nutrient delivery and the health of these vital ecosystems. To investigate these small-scale relationships, we deployed an uncrewed surface vessel (USV) to conduct a series of surveys around different kelp canopy densities. The USV was equipped with an acoustic Doppler current profiler, to measure water velocities, and a conductivity, temperature, and depth profiler, to measure physical properties of the water over time. These initial surveys revealed horizontal temperature gradients in and out of the kelp forest. Slight salinity gradients were also observed. This suggests a separation of water masses between waters in the kelp forest and surrounding it, likely attributed to waters inside the kelp forest residing in the area longer than surrounding open waters. From this study, we have a better understanding of how kelp forests can alter the physical structure of the nearshore water column and influence nutrient delivery and retention in biodiverse coastal systems.

Faculty Mentor: Dr. Trevor W. Hayton | Chemistry
Graduate Mentor: Gabriella S. Godinho
Additional Faculty Mentors: Dr. Justin J. Wilson and Kevin K. Lee | Chemistry

Presentation Title:
Uranyl Coordination with Dibenzo-diaza-18-crown-6 Macrocycle

Finding ways to isolate and utilize byproducts of energy and nuclear fuel production, such as the uranyl ion, has long been of interest to scientists. Our project involves coordinating uranyl with a macrocyclic ligand. Our particular interest in macrocyclic compounds, first synthesized by Curtis and House in 1961, arises from the fact that they are polydentate, chelating ligands constrained within large rings. It would be interesting to investigate how uranyl, with its strong preference to stay in linear conformation, interacts with macrocycles. The ligand we investigated, dibenzo-diaza-18-crown-6 Macrocycle, abbreviated as DADBC, successfully coordinated with the uranyl ion. By reacting UO2(OTf)2(THF)3 with 1 equiv. of DADBC, uranyl was successfully “threaded” into the binding cavity of the macrocycle, coordinated by dative bonds from two N and four O donor atoms occupying all of the equatorial positions. The resulting deep orange crystals were obtained by solvent extraction and crystallization via Et2O/MeCN layering. The identity of this complex was confirmed via 1H and 13C NMR along with X-Ray crystallography, UV-vis, and Raman Spectrospy. The complex exhibits C2 symmetry point group in both solid and solution states. The success of this project provides insight into the nature of uranyl ligand coordination, its binding tendencies, and what factors are considered for favorable and unfavorable complexation.

Faculty Mentor: Dr. Joe Blankholm | Religious Studies
Graduate Mentors: Shakir Stephen

Presentation Title:
Body Talk: Investigating the Confluence of Health Discourse and Contemporary American Religiosity

Contemporary American health discourses have evidently proliferated within the larger social, cultural, political, as well as religious sphere, suggesting the evolution of a nascent quasi-metaphysical and political vernacular within both evangelical and spiritual communities. By employing a multi-sited ethnographic approach, this project intends to research manifestations of this observable confluence. The two target demographics consist of (1) individuals who identify with some form of (largely Protestant-orientated) libertarianism (e.g., “MAHA,” or “Medical Freedom” advocates/conservative leaning Christians who profess anti-establishment ideas), and (2) “New-Age” or spiritually orientated health and “wellness” practitioners (e.g., reiki healers, yoga practitioners, doulas), and is primarily interested in how such individuals conceptualize health within a religious or spiritual framework. My thesis affirms that this apparent confluence is situated within a larger American historical past, one which has arguably created a religious milieu through combinative approaches—and that “health” as a site of personal freedom, autonomy, and liberty, has long constituted itself within a religious vernacular. Discourse around health, rather unsurprisingly, often becomes enmeshed within discourse around morality, ontology, faith, and autonomy.

Faculty Mentor: Dr. Zoe Liberman | Psychological & Brain Sciences
Graduate Mentor: Chuyi Yang

Presentation Title:
Assuming the Best: How Children Give the Benefit of the Doubt

Teachers are an essential component of children’s social and cognitive development. Indeed, children rely on teachers, and positively evaluate teachers who teach fully compared to those who withhold information (e.g., Gweon & Asaba, 2018). Here, we were interested in whether children can give teachers the benefit of the doubt when their teaching is incomplete. That is, do children ascribe positive intentions to teachers who only teach some information? To test this, we recruited a sample of 62 children aged 3-12 who were randomly assigned to one of two conditions: Partial Teaching or Full Teaching. In all cases, participants heard a story about a toy with 4 hidden features, and saw one student (the “teacher”) teach information about the toy to their peer. Across conditions, the teacher taught about either 2 features (Partial Teaching) or 4 features (Full Teaching). We asked participants an open-ended question about why the teacher taught 2 or 4 features. Then, children were either asked whether the teacher had positive intentions (wanted to help the peer learn), or negative intentions (wanted to appear smart). Preliminary results suggest that children do give benefit of the doubt: across age and condition children expected the teacher to have positive intentions. Children were less likely to endorse negative intentions, though this may vary based on how many features the teacher taught. We will also discuss implications and future directions of varying the social relationship between the teacher and student.

Faculty Mentor: Dr. Brandon Woo | Psychological & Brain Sciences

Presentation Title:
Who Do Children Choose to Learn From and Why?

Abstract:

Children do not learn equally from all the adults around them. Instead, they appear to be active learners who weigh social cues when deciding who to trust as a source of information. We investigated how children determine who to learn from, and whether relationships between caregivers and teachers might influence how children think about potential teachers. We approached these questions in two ways. First, in a correlational study, we examined associations between educators’ reported closeness with caregivers and their reports of children’s social and emotional well-being, as well as attentiveness. Findings suggest that when educators report stronger relationships with caregivers, they also describe the children as happier. Second, in an experimental study, we adapted established selective trust methods to test whether children prefer to learn from ambiguous teaching agents introduced as part of their caregivers’ social network or as someone new and outside that network. Together, these studies aim to shed light on how children’s social learning networks may be shaped not only by their direct experiences, but also by the relationships between the important adults in their lives.

Faculty Mentor: Dr. Matt Richardson | Feminist Studies
Graduate Mentor: Simone Temple

Presentation Title:
On the Research and Study of The No-Place Place of Black Trans and Gender Nonconforming Girls

Black girlhood studies focus on cisgendered girls and trans studies focus on white bodies. This leaves Black trans and gender nonconforming girls out of the conversation, rendering them silenced and unheard in both communities. Neither Here Nor There: The No-Place Place of Black Trans and Gender Nonconforming Girls explores this absence of a place for Black trans and gender nonconforming girls to share their experiences. To dive deeper, Black intersex children and adults rarely have opportunities to share their voices. Hearing their stories should not be impossible. My presentation, “On the Research and Study of The No-Place Place of Black Trans and Gender Nonconforming Girls,” will discuss the lack of resources about Black intersex lives and how the research uplifts and embraces Black trans, intersex and gender nonconforming girls.

Faculty Mentor: Dr. Yang Hai | Chemistry
Graduate Mentor: Wenhui Xi

Faculty Mentor: Dr. Craig Hawker | Chemistry
Graduate Mentor: Henry Wang
Postdoc Mentor: Dr. Christopher Cueto

Presentation Title:
Surveying the Morphological Behavior of Discrete-block-Random Disperse Copolymers

Diblock copolymers - comprised of two compositionally distinct, covalently attached polymer segments - exhibit diverse self assembly behavior on the nanoscale that can be utilized in technologies such as lithographic patterning. Different variables such as dispersity, repeat unit size, monomer sequence, etc. can be used to control self-assembly and morphology. In the present work a series of high χ-low N polymers consisting of a discrete or disperse dimethylsiloxane block paired with a random acrylate block were synthesized, with the goal of understanding the effects of both chain and compositional dispersity on ordered nanostructure formation.

Faculty Mentor: Dr. Todd Oakley | EEMB
Graduate Mentor: Cheyenne McKinley
Postdoc Mentor: Dr. Cory Berger

Presentation Title:
Uncovering Biological Rhythms in Ostracods: A New Model for Temporal Studies

Biological rhythms play a crucial role in regulating individual physiology, behavior, and interactions with both their environment and other organisms. These rhythms are often driven by endogenous oscillators, which enable organisms to anticipate and prepare for predictable environmental changes before they occur. While circadian rhythms are well studied, our understanding of other biological rhythms —such as circatidal, circalunar, and circannual rhythms— is still developing. Marine systems have played a valuable role in advancing research on the complex relationship between solar and lunar timing mechanisms. Ostracods with bioluminescence offer an excellent model for studying biological rhythms due to their well-characterized, naturally occurring courtship displays, which follow a circalunidian rhythm. They can also be maintained in laboratory settings using carefully recreated habitats. In our lab, we keep both bioluminescent and non-bioluminescent ostracods in custom tanks. It is unknown if non-bioluminescent ostracods possess the same circalunidian rhythm that bioluminescent species do. We seek to answer this by conducting laboratory experiments in an undescribed species, Skogsbergia sp. To further describe this species, we performed detailed morphological dissections on its limbs to compare its traits with those of known species. Our initial comparisons suggest similarities with Skogsbergia lerneri (Kornicker, 1958), particularly in the arrangement of basal spines, setae, and the structure of the antennae and lamina. These findings lay the groundwork for more comprehensive analyses aimed at resolving the species identity and deepening our understanding of biological timing systems in ostracods.

Faculty Mentor: Dr. Christopher Bates | Materials
Graduate Mentor: Andrew Rhode

Presentation Title:
Slide-Ring Hydrogel Mechanics, Adhesion, and Friction

Hydrogels are cross-linked hydrophilic polymers that are swollen in water. These complex materials can demonstrate viscoelasticity and allow the diffusion of solutes and water within the network. These properties make hydrogels ideal soft materials for a range of applications in aqueous environments, from contact lenses to tissue engineering. Here, we studied a slide-ring crosslinked polyacrylamide gel which contained cyclodextrin threaded onto polyethylene glycol chains. Slide-ring hydrogels can withstand substantial stress, yet still can exhibit low friction, making them potentially useful components for cartilage-replacement materials. We sought to understand how network structure influences friction and adhesion of these complex materials. We screened over 25 unique formulations with microindentation measurements. A subset of diverse samples were selected for low-throughput sliding measurements. We showed that there is a correlation between the adhesion measured by microindentation and friction measured by linear reciprocating sliding with glass countersurfaces. We related these measured properties back to key variables in the formulations, namely the water content, crosslinker content, and ring frequency along the polyethylene glycol chains. Ultimately, the results of this optimized workflow serve as a basis to design slide-ring gels for desired properties, such as high stiffness and lubricity.

Faculty Mentor: Dr. Daniel Conroy-Beam | Psychological & Brain Sciences
Graduate Mentors: Ashley Coventry & Carlos Sosa-Colindres

In dating, choosing a good match requires forming accurate impressions of potential partners. Unfortunately, prior literature indicates that first impressions are only weakly accurate. However, impression accuracy increases with greater exposure and the use of structured communication aides. Here, we sought to examine the efficacy of new dating preference conversation aides in increasing date impression accuracy and the identification of good dating partners. After self-reporting their mate preferences, N = 137 participants attended two, 40-minute, opposite-gender dates: one with a partner chosen to match with the participant’s preferences and another with a randomly assigned partner. Participants were given conversation starter cards with the option of using them during the date. Afterwards, participants reported their impression of their partner's traits, degree of preference match, and their interest in meeting again. Card use was a significant positive moderator of impression accuracy. As expected, card use was associated with increased identification of a poor match in random dates. Card use was also negatively associated with self-reported and perceived future date interest, but had no association with interest received from partners. Overall, our data suggest that structured conversation aides can be useful for improving identification of poorly matched dating partners.

Faculty Mentor: Dr. Mahnoosh Alizadeh | Electrical and Computer Engineering
Graduate Mentor: Arghavan Zibaie & Yohan John

Presentation Title:
Exploring Data Driven Models in the Realm of Industrial Refrigeration

Developing data-driven models that forecast temperatures in industrial refrigeration systems can enable energy-saving strategies such as load shifting. According to the International Institute of Refrigeration, refrigeration systems account for approximately 20% of global electricity consumption as of 2023, underscoring the potential impact of improving operational efficiency. To explore underlying trends and seasonality, we first analyzed Q̇ (rate of heat transfer) data from a facility equipped with a cloud-based industrial operating system. The data, collected at 4-minute intervals over a year, revealed both daily and monthly trends via Seasonal-Trend decomposition using LOESS. These temporal patterns informed feature selection for our forecasting models. Three models were developed to predict evaporator temperature 4 minutes ahead in a single room with a single evaporator, using current system states and environmental conditions: a linear regression model (baseline), a Gated Recurrent Unit (GRU) neural network, and a Sparse Gaussian Process (Sparse GP) model. The GRU achieved the highest predictive accuracy with a test mean squared error (MSE) of 0.0035 and an R² score of 0.9963, closely followed by the Sparse GP (MSE: 0.0038, R²: 0.996). The linear regression model performed substantially worse (MSE: 0.0479, R²: 0.9137). While the GRU delivered the best performance, the Sparse GP required approximately half the runtime (587.58 seconds) compared to the GRU (1157.66 seconds), demonstrating a trade-off between computational efficiency and accuracy. Next steps involve expanding to a full facility with simplified compressor configurations, using additional features and a 10-minute prediction horizon.

Faculty Mentor: Dr. Sebastian Streichan | Physics
Postdoc Mentor: Dr. Susie Wopat

Presentation Title:
Taking a deep dive: Exploring the tissue mechanics of mesoderm ingression during zebrafish gastrulation

The cell signaling pathways that guide body plan formation during vertebrate gastrulation have been extensively studied and spatially defined. However, a global picture of the underlying mechanics and physics that shape the embryo via these cell signals remains undetermined. To address this question, we use zebrafish as a model to understand the physics of gastrulation, which is marked by dramatic changes in embryo shape that result from epiboly movements and the layering of the ectoderm, mesoderm, and endoderm. During epiboly, tissue cells expand toward the vegetal pole via an actomyosin contractile ring, fully covering the yolk cell. Simultaneously, the mesoderm ingresses towards the animal pole. While the mechanics of epiboly expansion have been explored, the global mechanics of mesoderm ingression remain unclear. Therefore, our objective is to investigate the role of myosin in tissue flow during mesoderm ingression. We first used immunofluorescence to directly assess the location of all active myosin in the embryo. We have paired this with in toto imaging to obtain a 3D projection of the whole zebrafish embryo. From these 3D images, we implemented a Blender-based tissue cartography pipeline to develop a 2D surface map projection for better visualization and easier quantitative analysis. These results reveal an increase in active myosin in the mesoderm, suggesting a tissue-specific flow. In the future, we plan to visualize changes in myosin localization during the early stages of mesoderm formation to define the symmetry-breaking events that regulate mesoderm ingression.

Faculty Mentor: Dr. Constance Penley | Film and Media Studies
Graduate Mentor: Casey Coffee

Presentation Title:
Consent, Community, and a summer of mentorship with Dr. Constance Penley

This research investigates cultures of consent across three spheres: the adult industry, communities of fanfiction writers, and academe. I propose that sex workers in the adult industry and creators/consumers of erotic fanfiction exhibit a more nuanced understanding of consent than those tasked with the administration of campus sexual safety. I begin this research by exploring how the porn community negotiates consent both formally, through performer contract negotiations, and informally, through on-set dialogue. Educational resources, like the Free Speech Coalition’s INSPIRE program for industry newcomers, inform these negotiations. I bring to the investigation of consent negotiation in fandom my participation in these communities as a writer and reader. I have found the discussions of consent on Archive of Our Own, the largest fanfiction archive on the Internet, exceptionally compelling. The Archive operates through“tagging,” a system of coding content in each story, including scenes of non-consensual acts, acts of dubious consent, and acts of consensual non-consent. I investigate the tagging system as a form of “informed consent,” both in the archive and the community discussions it inspires. By contrast, my exploration of the literature on consent in academe reveals conflicting and incoherent policies, informed by ever-changing Title IX rules, that emphasize the dangers of sexuality rather than the possibilities of informed sexual exploration. The lack of sexual literacy in academe as compared to that found in porn and fandom prevents a wholly realized culture of consent that allows for pleasure as well as safety.

Faculty Mentor: Dr. William Robinson | Sociology

Presentation Title:
Was it Ever About the Animals? Conservation Initiatives and Displacement in Northern Tanzania: Interviews With the Maasai

State land grabs and restricted access for conservation initiatives today continue to be framed as one of ecological stewardship while displacing Maasai and other indigenous groups in the process. In January 2024, for example, the Tanzanian government announced that the Ngorongoro Conservation Area will no longer permit human settlement, though Maasai have occupied the land for hundreds of years and still do (Oakland Institute 2024). Such conservation initiatives fundamentally serve as a form of resource control with the aim of capital development, rather than the preservation of the resources themselves. Drawing from the twelve semi-structured, in-depth interviews with Northern Tanzanian Maasai community members conducted between December 2024 - January 2025, this article examines how Tanzanian state land conservation initiatives reflect global capitalist and post-colonial state relations evidenced in accumulation by dispossession. Using both a materialist and postcolonial theoretical framework, Maasai interpretation of state enacted displacement could be understood as global capital accumulation and post-colonial state control. It is necessary to bring the Maasai experiences to light, as firsthand experiences may challenge and create nuance in existing literature on state-Maasai relations and state legislation and contribute to critical debates on the state’s role of conservation and the role of capitalist accumulation in the postcolonial state’s development. This research is necessary as Maasai displacement is not a relic isolated from the past; it is not only current but also urgent and their struggle is ongoing. Displacement and its threats will continue the demands of the Maasai are overlooked by capital interest.

Faculty Mentor: Dr. Tomoyuki Ichiba | Statistics
Graduate Mentors: Qijin Shi & Ka Lok Lam

Presentation Title:
Optimal Investing Strategy using Asset Pricing Models

Investors seek to maximize returns on every investment or equity purchase. This study investigates optimal investment strategies using asset pricing models. Modern financial analysts have enhanced these models by treating financial data—such as price changes and returns—as time-series and stochastic in nature. In other words, financial data are both temporally ordered and randomly determined. To better model this behavior, machine learning techniques have been employed to understand the randomness inherent in financial markets. In this paper, we use both monthly and daily data from the open-source French Data Library, which provides returns for various factors and portfolios. We apply the Fama-French five-factor models to estimate pricing error—defined as the difference between the theoretical value and market price of a portfolio. We extract features from this pricing error using two techniques: Signature methods and Generalized Autoregressive Conditional Heteroskedasticity (GARCH). These features are then used to train three classification models: a Convolutional Neural Network (CNN), XGBoost, and an Ensemble of the two. Our findings show that signature features help reduce loss in noisy, high-variance datasets, and that the XGBoost model using GARCH features yielded the most accurate, precise, and high-recall predictions.

Faculty Mentor: Dr. Arpit Gupta | Computer Science
Graduate Mentor: Jaber Daneshamooz

Presentation Title:
Multi-City Validation of Broadband Subscription Tier Methodology: A Replication Study Across 30 US Markets

Crowdsourced speed test measurements are increasingly used to assess broadband performance, but raw speed test data lacks critical context about users' subscription plans, leading to potentially misleading conclusions about network quality and accessibility. The goal of our research is to test the generalizability of the Broadband Subscription Tier (BST) methodology, a statistical approach that infers users' subscribed broadband plans from speed test measurements across a broader geographic scope than previously studied. To test our hypothesis that BST remains effective beyond major metropolitan areas, we applied unsupervised clustering algorithms to speed test data from 30 major US cities, expanding upon previous work that focused on four metropolitan areas. We leveraged city-level broadband plan information from prior research (SIGCOMM'23) to reduce noise, as ISPs offer more limited plan ranges within cities compared to their national offerings, and our methodology groups speed test results into distinct performance tiers corresponding to actual subscription plans available in each market. The results indicate that BST maintains high accuracy across diverse geographic and infrastructural contexts, confirming that the majority of speed tests originate from lower subscription tiers. Our findings support the conclusion that speed test contextualization is essential for accurate broadband performance assessment regardless of market size or characteristics. This study is important because it validates a critical tool for broadband research across a representative sample of US markets, providing policymakers and researchers with a reliable methodology for contextualizing speed test data when evaluating broadband accessibility, equity, and actual consumer experiences with internet services.

Faculty Mentor: Dr. Tomoyuki Ichiba | Statistics
Graduate Mentors: Qijin Shi & Ka Lok Lam

Presentation Title:
The Signature Transform of the Logistic Map

The path signature transform is a novel computational transform that preserves the underlying dynamical properties of complex, high dimensional systems and data streams. Chaotic dynamical systems are a classical, deterministic approach to modeling the behavior of physical and numerical behavior of systems. The signature transform, like the Fourier transform, may be used to analyze the behavior of particularly complex systems. This work attempts an analysis of the signature transform applied to the logistic map, and discusses the limitations of the use of the path signature transform in the analysis of discrete one-dimensional systems. Further it proposes an experimental design that includes the signature transform and signature kernel to compress and preprocess video sensor data for the classification of key frames and path closure in a simultaneous localization and mapping data pipeline.

Faculty Mentor: Dr. Arpit Gupta | Computer Science
Graduate Mentor: Jaber Daneshamooz

Presentation Title:
NetAgent: An Agentic Workflow Automation System for Realistic Network Traffic Generation

The goal of our research is to address the critical lack of high-quality labeled networking datasets that have hindered the real-world adoption of machine learning models in networking applications. This study is important because existing Machine Learning (ML) solutions developed in controlled lab settings fail to generalize in production environments, primarily due to insufficient realistic training data. Current network traffic generation approaches are limited by their inability to automate diverse application workflows, restricting dataset creation to a narrow set of networked applications and limiting model generalizability. To test our hypothesis that AI-driven workflow automation can accelerate dataset curation, we developed NetAgent, an AI agent system comprising two key components: a State Synthesis Agent chooses the most suitable next action or system state based on the app’s current environment and goal, and a Runtime Agent that performs real-time application interactions through automated actions (clicking, typing, scrolling). We integrated NetAgent with netReplica, an existing network condition emulation framework, to generate comprehensive datasets under diverse network conditions. The results indicate that NetAgent successfully automated workflow generation for 4 video streaming platforms, 2 video conferencing applications, and 20 complex interactive websites. Our approach enabled reproducible traffic generation across varied network conditions. Our findings support the conclusion that AI-driven workflow automation significantly accelerates networking dataset creation. The combination of NetAgent and netReplica provides a scalable foundation for developing production-ready ML models in networking, potentially enabling widespread adoption of AI-powered network operations (AIOps).