Stat Cafe - Somjit Roy
Bayesian Symbolic Trees for Structural Learning of Scientific Expressions
- Time: Tuesday 10/28/2025 from 11:10 AM to 12:25 PM
- Location: BLOC 448
Description
Uncovering governing equations and symbolic relationships lies at the core of scientific reasoning, underpinning discoveries across physics, chemistry, and materials science. Yet, contemporary symbolic discovery remains dominated by heuristic, data-intensive Scientific Machine Learning paradigms that falter under noise and lack principled uncertainty quantification. Guided by interpretable Statistical Artificial Intelligence, we develop a hierarchical Bayesian framework that encapsulates scientific laws as ensembles of symbolic trees. This probabilistic construct facilitates coherent uncertainty quantification through efficient MCMC inference, balances predictive accuracy with structural parsimony, and enjoys theoretical guarantees of near-minimax posterior concentration. Robust performance is showcased through interpretable learning of symbolic expressions across diverse scientific domains.
Our Speaker
Somjit Roy is pursuing his doctoral degree in Statistics at Texas A&M University, under the supervision of Dr. Bani K. Mallick and Dr. Debdeep Pati. His research focuses on Scientific Machine Learning, Approximate Bayesian Methods, and Bayesian Optimization. He holds a B.Sc. and M.Sc. in Statistics from the University of Calcutta, India.