Code

Open-Source Code

This page collects selected open-source research code accompanying my work in uncertainty quantification, model calibration, and model-form uncertainty. These repositories are primarily research codebases intended to support reproducible experiments, methodological comparisons, and follow-up work in computational science and scientific machine learning.

Most of the current repositories are hosted under the UQUH GitHub organization. They are best viewed as research implementations rather than general-purpose software packages, but each repository is organized so that readers can reproduce the main experiments and adapt the methods to related problems.

Selected Repositories

• Open-Source Code
• Selected Repositories
• SS-PPCA: Stochastic Subspace via Probabilistic Principal Component Analysis
• SO-BO-scale: Bayesian Optimization under Uncertainty
• SS-Bootstrap: Stochastic Subspace via Bootstrap
• Notes

SS-PPCA: Stochastic Subspace via Probabilistic Principal Component Analysis

SS-PPCA is a research implementation of a stochastic subspace methodology for characterizing model-form error in high-dimensional computational models. The core idea is to build a probabilistic representation of reduced subspaces so that structured model discrepancy can be characterized more efficiently.

• Use it when: you want a structured probabilistic method for characterizing model-form uncertainty in high-dimensional simulation outputs.
• Where it fits: computational mechanics and related simulation settings where deterministic reduced-order models miss systematic discrepancy.
• Environment: MATLAB R2023b or later, with the Statistics and Machine Learning Toolbox and the Optimization Toolbox.
• What it includes: example problems for parametric linear static settings and higher-dimensional model-error characterization, with results written automatically to a results/ directory.

SO-BO-scale: Bayesian Optimization under Uncertainty

SO-BO-scale implements a Bayesian optimization framework for tuning a scale parameter in stochastic models under noisy evaluations. The repository is designed around calibration problems where direct trial-and-error tuning is expensive, unstable, or difficult to reproduce.

• Use it when: you need to calibrate a stochastic model with noisy objective evaluations and want a more systematic alternative to manual search or Monte Carlo-heavy tuning.
• Where it fits: uncertainty-aware model calibration, especially for stochastic subspace models and related scientific modeling workflows.
• Environment: MATLAB R2023b or later plus Python packages including numpy, scipy, matplotlib.
• What it includes: three example problems, the proposed BO implementation, GP-based and Monte Carlo baselines, state-of-the-art noisy-BO baselines, and a robustness study across seeds.

SS-Bootstrap: Stochastic Subspace via Bootstrap

SS-Bootstrap is a nonparametric stochastic subspace implementation for model-error characterization. Instead of relying on a parametric subspace model, it uses the empirical data distribution directly, making it a useful complement to SS-PPCA.

• Use it when: you want a nonparametric alternative for characterizing model-form error and prefer to work directly with the empirical distribution.
• Where it fits: stochastic reduced-order modeling and model-discrepancy analysis when parametric assumptions are undesirable or too restrictive.
• Environment: MATLAB R2023b or later, with the Statistics and Machine Learning Toolbox and the Optimization Toolbox.
• What it includes: an example benchmark problem, scripts for running the method end to end, and automatic saving of generated results and figures.

Notes

• These repositories are tied closely to the corresponding papers, so the best starting point is usually the README together with the paper or preprint.
• As the software portfolio grows, I plan to add more reusable research infrastructure and uncertainty-aware scientific AI tooling here.

For related papers and talks, see Publications and Talks.