Machine Learning Supporting Experimental Design for Product Development in the Lab

An interactive decision support framework is presented that assists lab researchers in finding optimal product recipes. Within this framework, an approach for sequential experimental design for black box models in a multicriteria optimization context is introduced. An additional criterion involving the prediction error to design new experiments is used with the goal to get a reliable estimate of the Pareto frontier within a few experimental iterations. The resulting decision support approach accompanies the chemist through the whole workflow and supports the user via interactive, graphical elements.