Machine Learning for Adaptive Experimental Design 

Machine learning identifies improved formulations up to 10 times quicker than traditional approaches, by focusing experimental effort directly on formulations that will lead to successful products in as few experimental cycles as possible

Identifying the optimal composition and processing parameters to achieve commercial performance goals as quickly as possible is the key objective of formulation design projects. Traditional approaches to formulation design suffer from key disadvantages: expert-driven design is labour-intensive and time-consuming; single-factor analysis misses the effects of correlations between factors, and conventional design of experiments is exhaustive but focused on covering the design space rather than rapidly achieving performance goals. Machine learning identifies improved formulations up to 10 times quicker than traditional approaches, by focusing experimental effort directly on formulations that will lead to successful products in as few experimental cycles as possible, adapting to the results of earlier measurements to ensure maximum value is extracted from the experimental outlay.