Parallel predictive entropy search for multi-objective Bayesian optimization with constraints applied to the tuning of machine learning algorithms

Eduardo C. Garrido-Merchán, Daniel Fernández-Sánchez, Daniel Hernández-LobatoExpert Systems with Applications 2023 · DOI · OpenReview

Abstract

Highlights•Bayesian optimization (BO) is the state-of-the-art in ML hyper-parameter tuning.•PPESMOC is a parallel BO method for constrained multi-objective optimization.•The chosen points to be evaluated in parallel reduce the most the solution’s entropy.•PPESMOC outperforms or matches other baselines in several optimization problems.•PPESMOC scales better with the batch size, enabling the use of bigger batches.
BibTeX
@article{garridomerchan2023parallel,
  author = {Eduardo C. Garrido-Merchán and Daniel Fernández-Sánchez and Daniel Hernández-Lobato},
  title = {{Parallel predictive entropy search for multi-objective Bayesian optimization with constraints applied to the tuning of machine learning algorithms}},
  journal = {Expert Systems with Applications},
  year = {2023},
  doi = {10.1016/j.eswa.2022.119328},
}
Figure