Leadership Alliance Summer Research Projects

Housing Justice Atlas (Dr. Brown Wilson): This Housing Justice Atlas project investigates immediate risk and predicts long term trends in housing instability. Working with community partners, we are co-creating data platforms for access to data, visualizations, and analysis to mobilize and empower impacted communities to reduce housing instability. The focal point of this work is the creation of a predictive model to examine the scope of the eviction challenges under different policy scenarios as COVID-19 eviction restrictions are lifted. By convening advocates and residents across both regions to evaluate the model, we will ensure the work addresses the central questions of stakeholders and integrates community expertise in the research effort. The pilot work in Charlottesville and Richmond will help build a framework to be expanded across the state as we assess the data sources and methods for scalability in other regions.

Climate Justice Mapping (Dr. Claibourn): The primary goal of the Climate Justice Mapping project is to build platforms for the collaborative identification, collection, and dissemination and analysis of information about the disproportionate harm of environmental injustice. We are co-producing a series of accessible, interactive climate justice maps, graphics, and models to  explore and visually represent local connections between systemic racism, environmental justice, and the local disparate effects of climate change. The project will also evaluate methods for integrating different forms of data captured at various spatial levels to improve estimation. This work will be used to raise awareness in equitable decision-making and to identify specific opportunities for equitable climate planning.

Mobile sensing, machine learning, health.  Research Project:  Modeling human behavior from mobile data streams

Our research team produces automated analysis techniques to make complex systems more dependable.  The Leadership Alliance (TLA) and NSF Research Experiences for Undergraduates (REU) students that join us will have the opportunity to learn about such techniques, contribute to their development, and apply them to autonomous systems like self-driving cars and swarms of drones.

Assuring that software in autonomous systems works correctly is critical for the health and well-being of people interacting with such systems.  Autonomous systems use machine learned (ML) software components and our lab performs research on how to validate that  they work as intended.  This involves exploring new ways to describe what an ML model should do (aka “specification”), to run an ML model on a systematically selected set of inputs to check its behavior (aka “testing”), and to conduct more thorough evaluation of ML models to guarantee that they behave correctly (aka “verification”).   Students will learn about ML models, apply state-of-the-art research tools to reason about them, and work with a team of PhD students on a project to advance the state-of-the-art.

High performance computing,  accelerating diverse applications by mapping algorithms and/or designing new algorithms for current accelerator hardware (such as GPUs), and designing novel accelerators.  

We have projects on using existing hardware accelerators and also designing novel accelerators for diverse applications, such as in astronomy, bioinformatics, and network processing.

The UVA NLP group currently focuses on advancing language technologies for facilitating human-AI collaboration. Specifically, we are working on (1) building explainable and robust deep learning models for language understanding and (2) designing learning algorithms to generate coherent natural language texts. Both topics will eventually be contributing to building AI systems that can easily collaborate with human users for certain language-related tasks.

Each student in this group has a project led by herself/himself. On top of that, we highly encourage students to collaborate with each other for making research progress. You can find the current team members via the following link: https://uvanlp.org/team/

You can talk to me to get more details if you are interested in one of the following topics:

1. Natural language generation for online story writing
2. Explainable deep learning models for long-document classification

Recommender systems, algorithmic explanation of personalized systems, network embedding, reinforcement learning

Research Projects include:  building metadata search engine, explainable recommendation, learning to rank
 

Our research is focused on understanding and mitigating risks associated with machine learning, especially when it is used in adversarial contexts. We do a mix of theoretical and experimental work. Experience with programming (mostly in Python), statistics, and machine learning are very helpful, but we have had undergraduate (and high school) student researchers in our group with little previous experience.  See https://uvasrg.github.io/ for examples of the projects we are working on.

AI for Translational Science (Dr. Brown): The integrated Translational Health Research Institute of Virginia (iTHRIV) is a federally funded collaboration between UVA, Virginia Tech, Carilion Clinic, and Inova that facilitates clinical and translational research across the Commonwealth. With partner institutions across the country researchers in iTHRIV have participated in creation of the National COVID Cohort Collaborative which now contains records from more than 5M patients that are helping to answer questions critical to understanding the pandemic. Students interns will learn about electronic health records and data privacy, data engineering/integration, data mining and big data analytics, and the use of artificial intelligence tools to help discover patterns of interest in the data.
 

Health Informatics (Dr. Valdez): Most illness severity scoring systems and continuous risk assessment strategies use a single mathematical model to relate measured parameters to the risk of an event, without attending to relevant individual characteristics. Our team’s previous research has demonstrated that disease progression and age influence clinical and subclinical processes for cardiorespiratory decompensation. For example, a model that performs well for someone in the early stages of sepsis does not perform as well for an individual with heart failure exacerbation. The same maybe true for individuals whose experience of health disparity has altered their baseline physiology, for example, by poorly treated hypertension. Clinical research has further shown variation in relevant risk profiles across other demographic characteristics. As a result, our current work focuses on the ways in which sex as a biological variable and variables associated with health disparities (i.e. race, ethnicity, geographic location, socioeconomic status, disability status, sexual orientation and gender identity) impact clinical and subclinical processes, allowing for individualization of such predictive models for cardiorespiratory decompensation. Students involved in this project will be mentored in both the social and technical components of creating and evaluating the performance of such tailored predictive algorithms.

Generalized epidemic modeling framework in Python (GEMFPy) (Dr. Shakeri): In this project, we are working on a Python simulator for stochastic spreading processes over networks. Students will learn how to speed up their codes and have the opportunity to contribute to a popular Python package. Previous knowledge of computer algorithms, Python programming is necessary and familiarity with linear algebra will be helpful.

Graph algorithm (Dr. Shakeri): Finding the shortest cycle in graphs, whose length is also known as the girth of the graph, is a fundamental problem in graph theory. Algorithmically, this is in general not as efficient as finding shortest paths. An efficient algorithm to find shortest cycles of a graph has applications to the determination of minimal cycle basis and maximum cycle packing. In particular, we encountered the need for such an efficient algorithm when studying the richness of cycle families for clustering algorithms in unsupervised machine learning. The shortest cycle problem is also related to graph properties such as chromatic numbers and connectivity. Furthermore, for planar graphs it corresponds to the min-cut problem in the dual graph. Students will learn about algorithm development.

 Opioid Epidemics (Dr. Shakeri): In this project, we aim to enhance the currently available methods to monitor opioid epidemics with several new sources. First, we use existing data from sewersheds (i.e. metabolite levels) that have been shown to provide more targeted and timely measures of prevalence for local sub-populations. We will also incorporate electronic health records, pharmacy data, etc. We will develop new statistical techniques to predict infection and recovery from pooled (spatially interval-censored) sewershed data, develop advanced multi-resolution forecasting models by combining all sources of data and estimates within our existing epidemiological modeling framework and provide tools for scenario planning, targeted alerts, and optimal decision-making. Students will learn about data integration, news and event monitoring, spatial analysis and predictive modelling.

Educational Analytics (Dr. Wright): This project will build on a recently created clinically integrated network (CIN) to (1) operationalize indicators of adolescent mental health treatment and (2) use data science to (a) examine patterns and prevalence of adolescents’ mental health treatment throughout the COVID-19 pandemic and (b) differences by Medicaid eligibility in Albemarle County, Virginia. Furthermore, we will use a new school indicator in the CIN to (3) examine how the schedule of remote instruction relates to income-based disparities among adolescents in Albemarle County Public Schools. Altogether, the project will lay the foundation to apply for a larger grant to integrate school and medical record data to examine income- based disparities in mental health and other indicators of psychological well-being across the COVID- 19 pandemic.

Mapping Surveillance Tech (Ms. Cummings) With the computational power, access to Big Data and support of Big Tech, law enforcement is building an algorithmic arsenal of AI firepower. AI policing is destined to become a superpower. But algorithmic abuse, through AI policing tools and AI policing tactics, could destroy lives in real time. Our data is not only being monetized as it is being shared with a broad range of law enforcement agencies; our data is also being weaponized against us. The brutal algorithmic force, through AI policing and AI surveillance, must be reined in, the overreach is scary and deadly. But most citizens are unaware of the power and pervasiveness of AI policing and the ubiquitous algorithmic information gathering strategies of law enforcement. AI has upgraded the undercover agent; algorithms are the new detectives, undercover officers and informants; taking covert policing to an unimaginable level. But the legal hurdles are way too low and the ethical guardrails way too weak for the encroachment and penetration of AI policing. We are being algorithmically cornered by the police in this Big Data Policing reality. This project culminates in the design of an interactive risk rating system of surveillance technologies to measure the intensity of algorithmic force deployed by the police. It strives to uphold due process, protect civil rights, and empower citizens with a tool that demands accountability and transparency in policing. It reimagines policing within a data justice and social justice framework.

Decriminalization (Dr. Kropko) We examine the complete dataset of court records in Virginia from 2000 through 2020 to measure the extent to which criminal record expungements under a new law expanding expungement eligibility are available to different segments of the population of Virginia, with a specific emphasis on race-based disparities. We also examine whether probation is applied in an unequal fashion across these groups.

Decriminalization of Homelessness (Dr. Alonzi) This project aims to evaluate the mental health crisis response system in Charlottesville and the surrounding community. This work is in alignment with the CIT Regional Coordinator and the mission of the Charlottesville Are Community Foundation (CACF) to "improve the quality of life for those living and working in the city of Charlottesville and the counties of Albemarle, Buckingham, Fluvanna, Greene, Louisa, Nelson, and Orange." This project will build upon the work of the existing partnership between community stakeholders and the University of Virginia funded by CACF. This interdisciplinary project brings together The School of Data Science, The School of Engineering, the UVA Health System, and iTHRIV at the University of Virginia. Community partners include: the regional jail, the local community service board, the police department, the local coalition for the homeless, and OAR JACC.

Understanding Misinformation as a Spreading Process (Dr. Henry) Understanding how misinformation, be it fake news stories, conspiracy theories, or incorrect scientific information, spreads through the social media landscape and changes as it spreads is a key challenge in building effective intervention strategies. A difficulty with studying misinformation has been in data availability, however recently a large database has been made publicly available called FakeNewsNet (link), which includes both content data on fake and real news stories across a number of domains, as well as data on the social media spread of these news stories. This capstone will focus on analyzing the types of fake and real news stories as a more granular level than is typically done, ideally through the use of natural language processing methods applied to the news story text. Additionally, time permitting, understanding how the different characteristics of the real/fake news stories predicts their spread through social media is of great interest.

Urban Data for the Night (Dr. Reia) Nightlife is a crucial aspect of our contemporary cities, encompassing various territories, communities, and economic activities. It has emerged as a vital component of urban governance worldwide, which led to policies to contain, support, and develop the urban night. Conflicts over the “right to the night” involving several stakeholders have been addressed differently by municipalities, from the appointment of “Night Mayors” (Amsterdam, Montreal, London, San Luis Potosi), Night Councils (Paris, Nantes, Montreal) and Night Offices (NYC, Washington D.C., Toronto). The increasing datafication of urban spaces has impacted the city after dark, even if this aspect of urban life is rarely the focus of data-centric initiatives beyond lighting and mobility. While certain cities rely heavily on big data to manage their territories, the lack of open/public datasets to comprehend the night is still an issue that leaves governments, businesses owners, activists, and night-shifters navigating challenging circumstances without essential information. At the same time, the night is inhabited by marginalized communities that might be harmed by increased visibility (and available data), shedding light on issues related to precarious labor, legal status, gender identities and expressions, and the limits of regulatory efforts. The pressure to measure nightlife grew with the Covid-19 pandemic, urging us to create mechanisms to use big data with a robust ethical approach that prioritizes accountability, transparency, and social justice. The research team in this project will take up the challenge of responsibly mapping, designing and publicizing datasets for measuring the formal and informal nighttime economies in three major North American cities: Washington, D.C., Montreal, and Mexico City. Our priority areas are Cultural venues; Labour; and Climate Justice.

Graph Unlearning for the Right-to-be-Forgotten (Dr. Wang) The right to be forgotten states that a data subject has the right to erase their data from an entity storing it. In the context of machine learning (ML), it requires the ML model provider to remove the data subject's data from the training set used to build the ML model, a process known as machine unlearning. While straightforward and legitimate, retraining the ML model from scratch upon receiving unlearning requests incurs a high computational overhead when the training set is large. To address this issue, a number of approximate algorithms have been proposed in the domain of image and text data, among which a popular method randomly partitions the training set into multiple shards and trains a constituent model for each shard. However, directly applying this to the graph data can severely damage the graph structural information, and thereby the resulting ML model utility.  In this project, we study how to apply machine unlearning to graph neural networks.  One idea is to consider new graph partition algorithms and a learning-based aggregation method. We will conduct extensive experiments on real-world graph datasets to illustrate the unlearning efficiency and model utility of our proposals.

The High-Performance Low-Power (HPLP) Laboratory is dedicated to research in the areas of Artificial Intelligence (AI) hardware and edge IoT. Ongoing research spans a wide range of topics from Processing-in-Memory (PiM), low-power hardware accelerator design, and Smart Dust, to explorations in spintronics and nanoelectronics.

Research Projects include:  
AI RISC - AI ISA extensions to RISC V
Smart dust - miniaturized sensor IoT nodes
Pimulator - FPGA emulator for PiM 
 

I lead the Collaborative Robotics Lab (CRL) at UVA. My group is interested in building robotic systems capable of working alongside people in complex human environments. In this project, the students will be involved in the project in various ways, from reviewing the literature to developing algorithms to implementing those on real robots to perform validation studies. Research to include close proximity human-robot collaboration in factory environments.

The Boukhechba group develops novel ubiquitous sensing methods such as wearables and ambient sensing tools to understand human behaviors. The group fuses advanced sensor signal processing, data analytics and machine learning methods to design novel sensing platforms used in multiple applications such as health (e.g., depression, anxiety, cancer, infectious disease and traumatic brain injury), and wellbeing (e.g., sports analytics, contact tracing). Students will learn how to design novel tools to understand human behaviors from ubiquitous devices such as mobile and ambient sensors. Students will closely work with other graduate students and postdocs from Dr. Boukhechba’s lab to design and test the next generation of ubiquitous sensing methods. Students will participate in all phases of research including literature review, designing prototypes and pilot studies, and writing papers.

Artificial Intelligence guided Autonomous Materials Discovery

We are an interdisciplinary research group interested in combining artificial intelligence (AI) with problems in materials science and engineering (MSE). Discovering new materials with targeted properties is a non-trivial task. Our group focuses on developing innovative AI-driven methods to accelerate the search and discovery of new materials. Of particular interest in this research project will be to develop AI-based tools that will autonomously guide computations to discover new, rare-earth-free, sustainable magnets that will find application in quantum information technologies.

Bio-inspired fluid dynamics of flying and swimming in nature. Students will learn basic skills in processing high-speed images, modeling, and flow simulation and visualization. Basic methodologies include 3D modeling software, e.g. Autodesk Maya®, numerical simulations of flows using in-house/commercial computational fluid dynamics (CFD) solvers, e.g. PICAR, ANSYS  Fluent®, and graphing tools, e.g. Tecplot® etc. Working with experimentalist and biologist, we will study the fluid dynamics and aerodynamics of a flying snake.

More efficient catalytic processes offer the opportunity to reduce environmental impact and energy consumption. The Gunnoe group focused on fundamental studies of catalysts for the activation and conversion of small molecules. Our interests are directed toward processes of relevance to the energy sector as well as large chemical processes, including more efficient conversion of light alkanes from natural gas, new strategies for the synthesis of alkyl and alkenyl areas, and catalytic processes relevant to scalable conversion of solar energy to chemical fuels. 

We can accommodate undergraduate students on any of our projects: 1. Electrochemical water oxidation (relevant to the conversion of solar energy to chemical fuels). 2. Catalytic synthesis of alkyl and alkenyl arenas (more energy efficient production of large-scale chemicals that are precursors for plastics, detergents and pharmaceutical/agricultural products). 3. Catalytic CH activation and functionalization for the conversion of light alkanes from natural gas into high value chemicals. 

We are a psychology lab at the University of Virginia, led by Dr. Noelle Hurd. We  focus on finding ways to promote the positive development of youth. Specifically, our focus is identifying ways to build up pre-existing strengths in youths’ lives. We work within both the university and the larger Charlottesville communities to conduct research and learn more about the factors that lead to healthy development. Methodologies include survey research, as well as the analysis of quantitative and qualitative data.  

We have a research project focused on developing a white bystander intervention targeting college students and addressing online racial discrimination.

We study the cognitive and social processes underlying communication and social interaction to better understand and support autistic individuals and their families. We want our work to be relevant and useful to the community, and believe that this is best accomplished when the people affected by the research collaborate in it. We envision a world where all individuals are welcomed, included, supported, and valued.

In these pandemic times, our methodologies involve primarily online cognitive and social psychology experiments, large survey analyses, coding of videotaped interactions, etc. Students will have a chance to work closely with members of the research team to develop skills in online experiment preparation, data analysis, and outreach and collaborative research with members of the autism community.

Research project to be determined,  however, several projects that will be in the midst of online data collection and/or in the analysis phase. Some address questions about emotion perception in autism, others involve questions about stigma and autism, etc.

Ribic lab studies brain connectivity, specifically how molecules that build neuronal connections guide functional development and learning. We use electrophysiology and 2-photon imaging in awake mice, in combination with genetic and viral tools.  

Different types of projects are available, and all focus on using mice as a model system.

We study social psychological processes that contribute to social disparities such as how racial bias in perceptions of others’ pain contributes to racial disparities in healthcare, how socioeconomic differences in access to and use of public space contributes to socioeconomic disparities in education, how gender differences in felt safety contributes to gender gaps in academia.  Students in our lab employ a variety of methods including surveys, experiments, and archival studies.  

The Biocomplexity Institute's Social and Decision Analytics Division (SDAD) and the School of Data Science at University of Virginia are seeking applications for graduate fellows and undergraduate interns for the Data Science for the Public Good (DSPG) Young Scholars program. Fellows and interns will work in teams collaborating with postdoctoral associates and research faculty, and project stakeholders. The research teams will combine disciplines including statistics, data science, and social and behavioral sciences to address complex problems proposed by local, state, and federal government sponsors and community stakeholders. The program teaches students how to discover and integrate vast amounts of data to study topics such as public health, education, public safety and economic mobility.  

The projects are policy-focused public good issues raised by local, state, and federal government sponsors and community stakeholders. The program teaches students how to discover and integrate vast amounts of data to study topics such as public health, education, public safety and economic mobility.

Projects from DSPG Young Scholars 2020 program can be found at: https://datascienceforthepublicgood.org/events/symposium2020/poster-sessions