2025
If available, a rigorous, first-principles model is preferable. Bayesian Optimization (BO) becomes interesting if such models are not available or too hard to develop or computationally prohibitive. BO is a data-efficient and easy-to-apply alternative. Its greatest strengths lie in its black-box nature, requiring no internal knowledge of the system, and its inherent robustness to observation noise; a critical advantage when dealing with real-world experimental data. The core of Bayesian Optimization is the concept of a surrogate model. The BO workflow consists of cycling between model updates and experimental evaluation. It is demonstrated using the Cinnamaldehyde (CHYD) Hydrogenation Model. By understanding its strengths and limitations relative to traditional methods, one can select the right tool. Two Python packages to facilitate demonstration are available: (i) bayesian-optimization-demo, and (ii) single-task-vs-multi-task.
