Data-Efficient Machine Learning for Smart Design of Experiments

The project has successfully enhanced Bayesian optimization methods, delivering solutions for both industrial and non-industrial applications. The technology has proven effective across diverse sectors, from industrial companies optimizing high-cost production processes and raw material usage, to logistics providers improving operations and supply chain management, and even healthcare simulation tool providers. Multiple state-of-the-art methods have been successfully implemented to handle mixed input spaces, significantly expanding the technology's practical applications. This project has completed the development of an advanced simulation model that now accommodates continuous, ordinal, categorical, and conditional decision variables. Comprehensive benchmarking of these novel approaches has been conducted across various use cases, including both analytical test problems and simulator applications, with documented positive early results. The research is now ready for industry implementation in settings where traditional manufacturing processes are being replaced by novel technologies, offering a more efficient alternative to traditional trial-and-error experiments for process parameter optimization. The versatility of the solution has been demonstrated through its application in optimizing parameters for simulation tools across multiple fields, particularly benefiting scenarios involving high operating costs and high-value raw materials.

Downloads

ABOT toolbox: extension to handle mixed input spaces -  https://gitlab.ilabt.imec.be/srojasgonzalez/abot-toolbox-v2 

'Getting started’ guide

Contact

Philippe Leroux (UGent) and Simon Appeltans (UHasselt)
Researchers: Ivo Couckuyt (PI), Inneke Van Nieuwenhuyse (PI) and Jeroen Jordens (Senior research engineer)
Research groups: UGent – imec (IDLab), UHasselt - Research Group Computational Mathematics, Flanders Make - CodesignS