{"id":40513,"date":"2024-11-22T07:00:40","date_gmt":"2024-11-22T01:30:40","guid":{"rendered":"https:\/\/www.globalindian.com\/youth\/?post_type=story&p=40513"},"modified":"2024-11-22T10:29:33","modified_gmt":"2024-11-22T04:59:33","slug":"achyuta-rajaram-regeneron-science-talent-search-winner-decoding-ai-decision-making","status":"publish","type":"story","link":"https:\/\/old.globalindian.com\/youth\/story\/cover-story\/achyuta-rajaram-regeneron-science-talent-search-winner-decoding-ai-decision-making\/","title":{"rendered":"Achyuta Rajaram: Regeneron Science Talent Search winner decoding AI decision-making"},"content":{"rendered":"

(November 22, 2024)<\/strong> Achyuta Rajaram, a 17-year-old senior at Phillips Exeter Academy, stood on stage in Washington, D.C., shocked as his name was announced as the top winner of the 2024 Regeneron Science Talent Search. His project, which developed a method to identify the decision-making components of artificial intelligence (AI) systems, earned him the competition\u2019s $250,000 prize. It was a moment of recognition for years of disciplined effort in understanding machine learning and pushing its boundaries.<\/p>\n

\u201cWhen they called my name, I honestly thought I might faint,\u201d Rajaram recalled. \u201cBy the time they announced the first runner-up, I had fully accepted that I wasn\u2019t in the top ten. I was genuinely happy for my friends and ready to celebrate their success.\u201d Instead, the recognition was his.<\/p>\n

The Regeneron Science Talent Search, often referred to as the \u201cJunior Nobel Prize,\u201d is one of the most prestigious STEM competitions in the United States. It challenges high school seniors to develop innovative projects while demonstrating academic versatility and depth. More than 2,000 students applied in 2024, competing in what organizers described as the largest entry pool since the 1960s.<\/p>\n

\"Achyuta

Addressing a key problem in AI research<\/strong><\/h4>\n

Rajaram\u2019s project addresses a key problem in AI research: understanding how models process information. While AI systems have become ubiquitous\u2014generating images, diagnosing diseases, or driving cars\u2014their decision-making processes often remain opaque. This lack of transparency creates risks, from unfair biases to dangerous misclassifications.<\/p>\n

Rajaram\u2019s method automates the identification of which parts of a model contribute to its decisions. \u201cYou can think of it like identifying which neurons in a human brain light up when you look at an image,\u201d he explained. \u201cExcept here, it\u2019s an algorithm, not a brain.\u201d<\/p>\n

This work is particularly relevant to image recognition models, which are used in applications ranging from facial recognition to autonomous vehicles. Rajaram tested his code on an open-source AI model and found that it helped resolve a significant flaw: misclassifying images when conflicting text appeared nearby. For instance, an image of a red traffic light labeled with the word \u201cgreen\u201d would sometimes be incorrectly identified as green. Rajaram\u2019s method pinpointed and addressed the problem by isolating the components responsible for the misclassification.<\/p>\n

\u201cAs AI models grow more complex, analyzing them manually becomes nearly impossible,\u201d he said. \u201cMy goal was to create a tool that could automate this process, making these systems more reliable and safer to use.\u201d<\/p>\n

Early Influences and Support<\/strong><\/h4>\n

\"Achyuta

Rajaram credits his interest in computer science to his parents, who both worked in software engineering. \u201cMy parents met while working at IBM in India,\u201d he said. \u201cFrom a young age, they encouraged me to be around computers and learn coding.\u201d This foundation gave him the confidence to dive into complex concepts and experiment with emerging technologies.<\/p>\n

At Phillips Exeter Academy, Rajaram quickly stood out for his academic curiosity and leadership. He co-leads the school\u2019s physics, chemistry, and chess clubs while maintaining a rigorous schedule. His interest in AI grew during his freshman year when he noticed that much of the research focused on natural language processing, such as chatbots, rather than image-based AI models.<\/p>\n

\u201cI saw a gap in the research and wanted to explore it further,\u201d he said. \u201cImages are central to so many AI applications, yet we don\u2019t fully understand how these models interpret visual data.\u201d<\/p>\n

Rajaram\u2019s project required more than curiosity\u2014it demanded time, discipline, and expert mentorship. He worked closely with Dr. Sarah Schwettmann at the Massachusetts Institute of Technology, spending three days a week in her lab. Each week, he dedicated 25 to 30 hours to refining his code and testing its applications.<\/p>\n

\u201cDr. Schwettmann\u2019s guidance was invaluable,\u201d he said. \u201cShe helped me think critically about the practical implications of my work and encouraged me to push the boundaries of what the project could achieve.\u201d<\/p>\n

\"\"