WORKSHOPS & TUTORIALS

AICOMP 2026 Workshop & Tutorial sessions

Bayesian Inverse Problems via Parallel Tempering

Trevor Campbell

University of British Columbia

Join our hands-on tutorial on nonreversible parallel tempering using the Pigeons platform (https://pigeons.run) for solving challenging, high-dimensional system identification problems. We’ll start by building a clear, intuitive understanding of how the method works—why running multiple interacting simulations at different settings helps explore possible solutions more effectively and avoid getting stuck in poor ones—before moving into practical use. From there, you’ll learn how to set up runs, interpret results, and improve reliability when standard optimization or sampling approaches struggle. You’ll leave with a solid grasp of the core ideas and the ability to apply them directly to your own engineering problems.

Uncertainty Aware Process Simulation

Goran Fernlund & Alberto Mussali

Convergent Manufacturing Technologies

Engineering models are used to describe physical phenomena and processes to better understand behavior and predict future outcomes. These models are typically deterministic, providing a single predicted response without indicating how reliable that prediction is. As a result, it is often difficult to assess confidence in simulation results or account for variability or uncertainty in model parameters or inputs.

Probabilistic modeling offers a complementary approach by explicitly representing both what is known and what is uncertain. By incorporating uncertainty into models, it becomes possible to quantify confidence in predictions, make better-informed decisions, and extract more value from limited test data.

This workshop introduces uncertainty-aware simulation for engineering applications, with a focus on process simulation. Topics include quantification of uncertainty in material and process models, construction of prediction and confidence intervals, and methods for assessing probabilistic model performance and reliability. Practical examples will illustrate how probabilistic models can be integrated into existing simulation workflows.

“We’re not replacing our models; we’re quantifying what they don’t know.”

Physics-Informed Machine Learning for Composites: From Processing Physics to Qualification and Certification

Navid Zobeiry

University of Washington

Machine learning has strong potential to accelerate the development and deployment of composite materials and structures. However, purely data-driven models are often difficult to generalize and may lack physical consistency. In composites, where data are typically limited and the governing physics span multiple length and time scales, effective use of machine learning requires closer integration with process modeling, mechanics, and uncertainty quantification. This tutorial presents a practical view of physics-informed and hybrid physics-machine learning approaches across the composite life cycle, from processing and material characterization to structural response, failure, and certification. The central idea is that the role of physics in machine learning is not the same across all problems. At the process level, physics can be incorporated directly through governing equations and constrained learning. At higher scales, where direct enforcement becomes less practical, physics can instead guide model structure, feature selection, transfer learning, and uncertainty propagation. Through several examples, this tutorial highlights how physics-integrated machine learning can reduce dependence on large datasets, improve interpretability, and provide a practical path for composite design, manufacturing, and digital certification.