Reliable Predictive Modeling Under Data, Fidelity & Hardware Constraints - Sahil Bhola's PhD defense

The methodologies developed in this thesis establish a framework for reliable predictive modeling under constraints on data, model fidelity, and computational resources. By addressing identifiability prior to inference, leveraging low-fidelity inductive bias in uncertainty-aware operator learning, rigorously quantifying uncertainty from reduced- precision arithmetic, and developing scalable approaches for finite-precision uncertainty propagation, this work advances the characterization, prediction, and inference of complex physical systems under practical resource limitations.