On Jan. 23, 2025, the Regeneron Pharmaceuticals and Society for Science announced the top 40 finalists for the 2025 Regeneron Science Talent Search, the US’s most prestigious and longest ongoing STEM competition. The competition has been going on for 84 years and asks high school seniors to come up with a solution for some of society’s most urgent problems. Some examples of past dilemmas that students came up with a solution for include improving automatic discovery of visual circuits, investigating the rapid growth of certain cancers, and many more.
DVHS senior Elisa Zhang qualified as one of the top 40 finalists for the competition. Zhang’s research was in the computer science category and her project was about repairing bugs in code, using large language model agents to aid in the process. Large language models are a form of AI, and they’re trained on all types of data. She fed it data specifically related to identifying bugs.
“I basically developed a framework for repairing multi-location bugs, and so basically what I did was I did it all based on a data set of Java bugs, which is the most commonly used benchmark data set right now. And basically, it was just trying to improve the repair accuracy of multi-location bugs compared to state-of-the-art, traditional repair tools,” Zhang said.
Zhang also mentioned how she constructed databases by using repair pattern information and other similar tools. “Then I provide that information to the large language model agents, and I provide the large language model agents with different prompts and I split up the tasks in order to carry out the bigger task.”
Zhang’s background in research stretches beyond this project, however. She had done research with George Mason University for two summers prior to starting the project, so she was already familiar with the research process.
“I started off with cybersecurity research,” Zhang shared. “So that research was mainly on code vulnerability detection, which is actually kind of related to repair because obviously, after detection comes repair.”
With this project, she decided to expand her research beyond cybersecurity towards general program repair as she did not want to confine herself to just cybersecurity. Since Zhang was already doing vulnerability detection, she felt like the next stop was to try and fix it.
“There’s a lot more research and benchmarks out there that are more for general program repair; there’s very limited data on vulnerability repair. And I wanted to focus on multi-location bugs because they’re really common in real world code,” Zhang said.
She spent the first six months trying to identify a research topic and approach, as well as preliminary testing of large language models. Afterwards, she began implementing and refining her design.
“I read a lot of papers, obviously. There was Google Scholar, and then there were some [papers from] top conferences, like ICSE and ACM,” Zhang said. “I also use a lot of Microsoft’s blogs because I use one of Microsoft’s frameworks to implement my code.”
Zhang has been to some conferences herself, and she talked about how the experience meeting graduate students and professors working in the same field motivated her to keep pursuing research.
“Everyone was really nice and welcoming to me, even though I was only a high schooler. And it is [surreal] how there’s all of these people who are experts in your field and pushing all these boundaries. It is cool to be able to contribute and come up with these new ideas that actually have an impact,” Zhang explained.
Zhang received guidance for the project from her research advisor, Kun Sun, a George Mason University professor, and her research mentor, Shiyu Sun, a PhD student at the same university.
“Fun fact, every Wednesday before school, I would meet with [Shiyu Sun] at 7:30 in the morning,” Zhang recalled. “The research and the information I was getting on my own, but the ideas and getting feedback on my three or four different possible approaches, that was what my mentor was there for.”
Shiyu Sun describes Zhang as confident, hardworking and a quick learner, explaining that while she helped Zhang with the research process, Zhang wrote her research paper and completed her STS application independently.
“She is the most insightful high school student I have ever mentored. She has an exceptional ability to grasp complex tasks with precision and complete them efficiently. She quickly understands new concepts, asks insightful questions and applies her knowledge effectively to solve problems,” Shiyu Sun said.
Kun Sun has known Zhang for the three years that she has been researching with George Mason University. He oversaw and approved Elisa’s project, providing general advice.
“I provided OpenAI support for her research experiments and met with her monthly to offer feedback,” he said. “The most rewarding part was witnessing her deepen her interest in STEM research and build a strong foundation for her future career.”
Zhang spent a year on her project before submitting it to STS. She credited her ability to complete her research to good time management, though she also highlighted the importance of maintaining balance in life and her gratitude towards her parent’s support.
“In high school, [my parents] were always really supportive of me doing whatever I liked. And so they didn’t push me to overload myself with APs or other extracurriculars or stuff like that. I guess that was how I had so much time to do my research,” Zhang said.
Zhang believes that a way that she was able to balance her project with school was also through her parents not forcing her to take a certain amount of classes during the school day. Since she only has five classes a year, she has extra time to do research for her project.
Zhang also dances throughout the year at Tri-Valley Dance Academy, something she noted has helped her keep this balance, especially when dealing with burnout.
“Dance is time for me to chill and forget about everything else that’s going on in my life and also time to have fun with friends, and so when I get stressed, I go to dance,” Zhang said. “Or sometimes I even just go to the studio by myself and I just practice. That’s the way I deal with burnout, it’s just taking a break through dance.”
Looking to the future, Zhang definitely wants to pursue computer science but expand beyond her research topic. Zhang hopes she will discover her specific CS-related interests in college.
“I enjoy CS, but I also think I’m more interested in applications of CS and AI, and stuff like that,” she said. “So, whatever I do in the future, I hope that it can contain multiple disciplines.”
