Somjit Roy, a third-year Ph.D. student in the Department of Statistics at Texas A&M University, has received the 2026 Student Paper Award from the American Statistical Association’s (ASA) Section on Bayesian Statistical Science. The award recognizes outstanding work by early-career researchers in Bayesian statistics.
Roy’s award-winning paper, “Hierarchical Bayesian Operator-induced Symbolic Regression Trees for Structural Learning of Scientific Expressions” introduces a novel approach to scientific machine learning aimed at making the discovery of scientific relationships more transparent and interpretable.
A native of Kolkata, India, Roy earned a Bachelor of Science in statistics from St. Xavier’s College (Autonomous), Kolkata, and later completed a Master of Science in statistics at the University of Calcutta. During his undergraduate and master’s studies, he developed a foundation in statistical modeling and probability, which sparked an interest in using statistics to address real-world scientific problems. That motivation led him to pursue doctoral research at Texas A&M, where he works with scientific machine learning and computational Bayesian statistics.
After completing his Ph.D., Roy hopes to pursue a research-focused career in either academia or industry, working at the frontiers of scientific discovery. He said his long-term goal is to develop statistical tools that move beyond accurate prediction and instead provide deeper mechanistic insight, helping domain experts better understand the laws and processes that shape complex scientific systems.
Roy said he did not expect to win. “I was not expecting this award because it’s quite a competitive process,” he said. "I'm happy to receive it and would like to thank my advisers and collaborator, Dr. Bani K. Mallick, Dr. Debdeep Pati, and Dr. Pritam Dey have encouraged me at every stage of my research.”
A Breakthrough in Scientific Discovery
Roy’s research is in the field of scientific machine learning, which combines scientific knowledge with modern computational methods to accelerate discovery. He works with symbolic regression, which aims to find mathematical formulas that describe scientific data.
Mallick emphasized that this approach is a meaningful step forward for scientific machine learning, “From my perspective, Somjit’s paper stands out because it moves scientific machine learning beyond black-box prediction toward principled scientific discovery.”
Roy described symbolic regression as a way to move toward identifying scientific relationships while quantifying uncertainty in scientific processes and experiments. “Symbolic regression is a powerful machinery that helps to discover important scientific relationships or mathematical relationships between your response and the independent variables,” he said. “This is especially relevant in materials science and engineering, where researchers try to identify algebraic combinations of physical features to explain underlying processes, known as descriptors. Machine learning including deep learning techniques that dominate the field of symbolic regression offers computational scalability, but it often focuses on prediction and lacks transparency, making it difficult to see what is happening inside the model or to quantify uncertainty,”
To connect his research to broader scientific challenges, Roy highlighted the limitations of traditional machine learning approaches. “Researchers across disciplines are trying to uncover the scientific laws that govern the data they collect,” he said. “While machine learning is great at prediction, it often functions as a black box, producing accurate results without revealing how those predictions are made or how confident we should be in them. This lack of transparency can limit their usefulness for scientific discovery.”
Roy’s interest in Bayesian statistics grew from its ability to incorporate scientific knowledge and represent uncertainty. “I have always been fascinated by the idea of Bayesian statistics, how it can quantify uncertainty,” he said, adding that it provides “the flexibility to incorporate scientific knowledge, to incorporate a variety of information that would be helpful to go to a robust set up or a robust modeling framework”.
The goal is not only to produce a formula, but to make the result interpretable to scientists and to create confidence in what is discovered. “This capability is valuable for academia because it accelerates data-driven learning and scientific understanding, and for society as it enables more efficient scientific exploration and experimentation. It can lower costs and speed up discovery in areas such as materials design, energy technologies, and engineering — helping to translate data into reliable and interpretable scientific knowledge.”
The Goals of a Statistician
The ASA Student Paper Award is regarded as a major honor in Bayesian statistics, with awardees selected from large numbers of submissions. As part of the recognition, Roy was invited to present at the Joint Statistical Meetings, which will take place this summer in Boston. It is described as “the world’s largest gathering of statisticians,” with more than 5,000 attendees from 52 countries.
Roy will also present his work in April at the Symposium on Data Science & Statistics in Milwaukee. The opportunities expand the visibility of his work and connect it with a broader community working across statistics, data science and scientific machine learning.
Mallick said Roy’s award reflects his individual achievements and ambition. “Somjit Roy is a highly talented and dedicated young researcher with exceptional potential,” he said. “His work demonstrates both strong theoretical foundation and a keen ability to develop innovative methodology for real-world scientific problems.”
Roy’s advice to other graduate students emphasized purpose over prizes. “Your goal should be solving problems,” he said, adding that researchers should work with a focus on the underlying problem and clarity in modeling. Mallick offered similar guidance: “Always begin by asking how the problem you are working on can make the world better,” he said.