Projects
Research Projects
Overview
- Uncertainty-aware Scientific Foundation Models — Probabilistic LoRA for better-calibrated foundation models. Read more
- Correcting Model-form Uncertainty — SROM corrections to reduce model error in safety-critical predictions. Read more
- Characterizing Model-form Uncertainty — SS-PPCA and SS-Bootstrap for characterizing model error. Read more
- Bayesian Optimization under Uncertainty — Faster hyperparameter tuning under noisy evaluations. Read more
- Uncertainty Quantification in PINNs — GAN-augmented PINNs for robust generalization. Read more
- Structural Health Monitoring using ABC — Probabilistic SHM under thermal variability. Read more
Uncertainty Quantification in Scientific Foundation Models
Duration: Jun 2025 — Present
Mentor: Dr. Ruda Zhang
Collaborator: Taiwo Adebiyi
Problem: Foundation models are typically fine-tuned deterministically, which limits uncertainty representation for scientific decision-making.
Approach: Probabilistic Low-Rank Adaptation (LoRA) strategies and Bayesian deep learning methods.
Tools: Python, PyTorch, LoRA.
Outcome: A probabilistic, better-calibrated model compared to deterministic baselines.
Correcting Model-form Uncertainty
Duration: Jun 2024 — Present
Mentor: Dr. Ruda Zhang
Problem: Structural discrepancies in computational models lead to unreliable predictions in safety-critical settings.
Approach: Stochastic reduced-order modeling (SROM) corrections at reduced dimensionality to lower computational cost.
Outcome: Improved predictive accuracy on benchmark computational mechanics problems.
Characterizing Model-form Uncertainty
Duration: Aug 2023 — Mar 2025
Mentor: Dr. Ruda Zhang
Problem: High-dimensional structural simulations are biased under incorrect model assumptions.
Approach: Stochastic Subspace via PPCA (SS-PPCA) and a bootstrap-based SS-Bootstrap method.
Outcome: Improved accuracy; presented at WCCM/PANACM 2024 and CASML 2024; SS-PPCA published in Computational Mechanics (Springer).
Bayesian Optimization under Uncertainty
Duration: Jul 2024 — Feb 2025
Mentor: Dr. Ruda Zhang
Problem: Noisy evaluations make hyperparameter optimization expensive.
Approach: Bayesian optimization applied to scale parameters in SS-PPCA and SS-Bootstrap.
Outcome: Reduced training data needs; up to 40x faster convergence vs. 1D Monte Carlo optimization.
Uncertainty Quantification in PINNs
Duration: Oct 2024 — Dec 2024
Institution: Rice University
Problem: PINNs often fail to generalize under uncertain inputs and boundary conditions.
Approach: GAN-augmented PINNs (inspired by Yang et al., 2019) to improve robustness.
Tools: PyTorch, GANs, PINNs.
Outcome: Improved robustness in uncertainty quantification tasks.
Structural Health Monitoring using ABC
Duration: Jul 2021 — Jun 2023
Mentor: Dr. Ananth Ramaswamy, IISc Bangalore
Problem: Thermal variability and nonlinear damage signatures complicate bridge monitoring.
Approach: Approximate Bayesian Computation (ABC) framework for SHM under varying temperature.
Outcome: Robust probabilistic framework; presented at ICCMS 2022.