useR! 2024

Bayesian Optimization has emerged as the de facto standard for optimizing computationally intensive black-box functions. Such functions, characterized by the lack of availability of any information beyond the output value for a given input, present significant challenges in domains ranging from hyperparameter optimization in machine learning to applied sciences such as chemical engineering, material sciences, and drug discovery. mlr3mbo offers a modern and versatile approach to Bayesian Optimization in R as part of the mlr3 ecosystem. It not only provides ready-to-use optimization algorithms but also provides the essential building blocks necessary for the easy development of custom Bayesian Optimization algorithms. This flexibility extends to supporting both single and multi-objective optimization problems, along with handling mixed search spaces that include continuous, categorical, and conditional variables. In this talk, we will showcase mlr3mbo and its key features with practical demonstrations on hyperparameter optimization in machine learning, illustrating its potential to boost efficiency and effectiveness.