PEA student wins $250K prize at Regeneron Science Talent Search
Have you ever used AI-powered software such as ChatGPT or Google Gemini to edit or even generate an image based on prompts? Have you ever wondered how the software is able to come up with accurate answers with little to no error? ● That’s what Phillips Exeter Academy student Achyuta Rajaram aimed to find out this past year, which earned him the top prize at the recent 2024 Regeneron Science Talent Search. The 17-year-old student won $250,000 for developing a code that discovers which part within AI software, particularly when working with images, does the “thinking” to obtain answers.
In simpler terms, Rajaram said his research is like cutting into the human brain and looking at which neurons are responsible for identifying pictures – but with AI software instead of the human body.
“I was honestly very, very shocked… one of the security guards said that I looked like I was about to faint,” he joked, referring to being called on stage as the top winner. “I was just extremely extraordinarily shocked by the announcement.”
Competing in the largest entry pool since the 1960s, with over 300 applicants from 19 differe-nt states, Rajaram did not expect to win. He said by the time they announced the first runner-up, he had “fully accepted that I didn’t make it to the top 10.”
“There’s sort of two worlds – either I do really well or one of my friends does really well. I was sort of fine with either,” he said.
With this prestigious award, Rajaram joins fellow PEA alumni including Neil Chowdry, who placed fifth in 2022, and Yunseo Kim, who placed first in 2021.
Rajaram was the second PEA student to take home the top prize at the Regeneron Science Talent Search, with Kim being the first.
A Hopedale, Massachusetts native, Rajaram came to PEA in ninth grade following a scholarship offer from the institution.
He is also the co-head of the physics, chemistry and chess clubs.
Rajaram said he has always had an interest in computer science and coding, citing exposure and encouragement from parents at a very young age.
“Both my parents have had a background in software engineering more broadly, I think they met each other when they were working at IBM back in India,” he said. “So they both encouraged me to be around computers and to learn to code things from a pretty young age.”
Rajaram's interest in this particular project began when he discovered that research in AI-based software has focused more on language instead of images.
As part of his research, Rajaram traveled to the Massachusetts Institute of Technology in Boston to work with Dr. Sarah Schwettmann three times a week, putting in anywhere between 25-30 hours to develop the code that would lead to his win.
According to the Society for Science, the organization behind the competition, Rajaram's research “improved our ability to discover what computer models… are ‘thinking' when they analyze a photo” and which parts of their “mechanical brains” are contributing to the decision making.
“Achyuta's (Rajatram) key contribution to this effort was to develop an automatic method for recognizing which parts of the algorithms identify what,” the organization stated. “This knowledge sheds light on what these algorithms are ‘thinking,' which can help make them more effective, fair, and safe.”
Just like many other teenagers, during his free time, Rajaram said he likes to hang out with his friends and eat, with a particular favorite being Szechuan Taste's honey spareribs.
After graduating from PEA this spring, Rajaram will attend MIT, where he plans to major in computer science and continue his research in AI. Looking ahead, Rajaram said the next step for this specific research is to use his codes on a wide range of open-source networks to see how broadly it is applicable.
“There's a lot of research in machine learning interpretability, and I really hope this is a tool in the larger toolbox to sort of better understand how these models (AI-based software) work,” he added.