From smartphone apps to hospital patient records, personal data are collected throughout daily life. These data can help organizations build better artificial intelligence systems, improve healthcare and inform public policy. Using and sharing them, however, can create serious privacy risks.
Tianhao Wang , an assistant professor of computer science at the University of Virginia School of Engineering and Applied Science, is developing ways to help organizations gain useful information from sensitive data while protecting the people represented in those data.
His work has earned him a National Science Foundation Faculty Early Career Development Program, or CAREER, Award. The program supports early-career faculty who show potential as role models in research and education.
Wang studies privacy-preserving AI and data analysis. “My research focuses on a simple but increasingly important question: ‘How can we use data to advance science and AI without exposing people’s private information?’” said Wang, who joined UVA in 2022 after earning his Ph.D. from Purdue University and completing a postdoctoral appointment at Carnegie Mellon. “The goal is to make more data safely usable for research and innovation.”
To do this, Wang’s research focuses on differential privacy — a mathematical framework that limits what an analysis or released dataset can reveal about any one person — and on privacy risks in AI systems, such as models memorizing or disclosing sensitive information from their training data.
The $677,866 CAREER Award for his project, “ Advancing Differentially Private Data Synthesis: A Holistic Approach ,” will support new methods for generating useful synthetic datasets while providing rigorous privacy protections.
“The most valuable datasets often contain sensitive information, such as medical records, financial transactions or personal images,” Wang said.
“My research aims to create realistic synthetic data — artificial data that preserves the useful patterns of the original dataset while protecting the privacy of the individuals who contributed the data.”
Wang is a key contributor to UVA Engineering’s breadth and depth in AI and cybersecurity research, said Sandhya Dwarkadas , Walter N. Munster Professor and chair of the Department of Computer Science.
“Today, we have access to vast amounts of data and compute power, with the potential to use it for society’s benefit,” Dwarkadas said, “Tianhao’s research will help unlock that potential by alleviating privacy concerns that would otherwise prevent those data from being shared.”
Synthetic datasets contain artificial records generated from patterns in real data, but they are not automatically private. A poorly designed AI system could memorize or reveal sensitive details. Wang’s project applies differential privacy during the generation process, which formally protects individual privacy by limiting how much any one person’s information can influence that learning process.
An artificial medical image, for example, should preserve characteristics that help researchers study diseases without copying or revealing the image of a real patient. Similarly, a synthetic table of patient records should reflect population-level trends without disclosing information about a specific individual.
“One way to think about it is that we are teaching an AI system the ‘big picture’ while preventing it from memorizing details about any one person,” Wang said.
The project also addresses another central challenge: the trade-off between privacy and data quality. Limiting what an AI model can learn about individuals can also make it more difficult to preserve all the information researchers need.
“Existing privacy-preserving methods can work reasonably well for simple data, but they still struggle with more complex data such as medical images and multimodal data that combine images, text, sensor readings and other information,” Wang explained.
His CAREER project focuses on three connected goals: preserving important statistical patterns and relationships in sensitive datasets; generating higher-quality synthetic images and multimodal data; and using public datasets and foundation models — large AI models trained on broad collections of data — to improve performance.
Together, these efforts will contribute to a broader framework for privacy-preserving data synthesis.
“Today, different types of data often require very different techniques,” Wang said. “We hope to identify common principles that can be applied across many domains.”
Although the project’s core methods come from computer science, determining whether synthetic data are useful requires knowledge from the field in which the data will be applied. Medical researchers, for example, may need synthetic data to preserve clinically meaningful relationships, while researchers studying public policy, finance or human behavior may have different requirements.
Wang hopes the research will create a foundation for future partnerships with researchers at UVA’s School of Medicine and other specialists who work with sensitive data.
“Computer scientists can develop the privacy and AI methods, but researchers in other fields can help us identify which parts of the data are scientifically important and how the synthetic data should be evaluated,” Wang said. “I look forward to collaborations that enable these methods to address real research needs, not only technical benchmarks.”
Such collaborations could help researchers evaluate whether synthetic data preserve the information needed for a particular scientific question while still meeting rigorous privacy requirements.
“The long-term vision is to make privacy-protected synthetic data generation as reliable and accessible as possible, so that organizations can responsibly unlock the value of sensitive data.”
As with all CAREER Awards, student training is central to the project. Graduate and undergraduate students are involved in developing algorithms, building software systems, running experiments and applying the methods to real-world fields such as healthcare.
Wang’s research also supports new hands-on course projects, open-source software tools and outreach activities designed to introduce students to privacy-preserving AI technologies.
Yucheng Fu, a second-year Ph.D. student in computer science, studies differential privacy and applied cryptography. He said Wang supports students in developing projects based on their own interests.
“That freedom and encouragement have made the research process especially rewarding,” Fu said. “Under Professor Wang’s guidance, I’ve had the opportunity to work on privacy-preserving computation and think about how rigorous privacy guarantees can be applied to real-world systems.”
Wang’s 2026 CAREER Award adds to UVA’s research leadership in cybersecurity, AI and data science. It also creates opportunities to connect advances in computer science with work in healthcare and other fields that need responsible ways to use sensitive data.
“Ultimately, I hope this work helps researchers, healthcare organizations, companies and government agencies safely share and use data that would otherwise remain inaccessible because of privacy concerns,” Wang said.