Scalable learning algorithms for process modeling

Learning from data is a key step in developing mathematical models for process simulation, optimization, and control. Data-driven mathematical models must satisfy the underlying principles of chemical engineering systems, such as conservation of mass and stability.

We develop control- and optimization-theory constrained learning frameworks and solution algorithms to learn interpretable mathematical models with guaranteed properties.

Some snippets of our previous work are presented below.

A constrained symbolic regression approach for Lyapunov function discovery

Learning interpretable and stable dynamical models via mixed-integer Lyapunov-constrained optimization

Discovering interpretable piecewise nonlinear model predictive control laws via symbolic decision trees

Learning regime-dependent governing equations: A symbolic decision tree approach

Selected Publications

Mitrai, I. and Tang, W., 2026. A constrained symbolic regression approach for Lyapunov function discovery. arXiv preprint arXiv:2606.10045. [arXiv]
Mitrai, I., Liu, T. and Sanoja, G.E., 2026. Learning regime-dependent governing equations: A symbolic decision tree approach. AIChE Journal, p.e70632. [paper] [arXiv]
Li, Z. and Mitrai, I., 2026. Learning interpretable and stable dynamical models via mixed-integer Lyapunov-constrained optimization. Accepted to 2026 Conference on Decision and Control, arXiv preprint arXiv:2604.07611. [arXiv]
Mitrai, I., 2025. Discovering interpretable piecewise nonlinear model predictive control laws via symbolic decision trees. In 2026 American Control Conference (ACC) (pp. 4248-4253). IEEE. [paper] [arXiv]

Page updated

Google Sites

Report abuse