OATML

Pritvik is a DPhil student in Computer Science at the University of Oxford under Prof. Yarin Gal’s OATML Group. His research lies at the intersection of machine learning and scientific discovery, with interests in continual learning, Bayesian methods, probabilistic and generative modelling, uncertainty quantification, and simulation-based inference. He is currently exploring Machine Learning Interatomic Potentials (MLIPs) and is particularly interested in developing reliable, interpretable models that can incorporate new information without catastrophic forgetting, remain calibrated under distribution shifts, quantify epistemic and aleatoric uncertainty, and recover latent physical structure from high-dimensional scientific data. His DPhil aims to move beyond static predictors trained on fixed datasets toward adaptive models that evolve alongside new data, simulations, and scientific knowledge. Before Oxford, Pritvik graduated from Caltech as a Rise Fellow with a BS in Physics and a minor in Astrophysics, where his research spanned cosmology, gravitational waves, particle physics, astrobiology, complexity theory, and machine learning, including work with the Caltech-CERN CMS Group under Prof. Harvey Newman, Caltech-LIGO Group under Prof. Alan Weinstein and Prof. Barry Barish, and Caltech-JPL collaborators, including Prof. Yuk L. Yung and Prof. Stuart Bartlett. He continues collaborations with UC Berkeley and Lawrence Berkeley National Laboratory on machine-learning approaches to primordial non-Gaussianity, supernova cosmology, and dark-energy inference. At OATML, Pritvik’s goal is to develop uncertainty-aware AI methods that expand what scientists can infer from complex simulations and next-generation datasets, using fundamental physics as both a scientific target and a rigorous testbed for new machine-learning methodology. Pritvik is also a fully funded Ellison Scholar for his doctoral studies at Oxford, and is co-supervised by Dr. Danilo Rezende at the Ellison Institute of Technology, Oxford.