Chemically-informed data-driven optimization (ChIDDO): leveraging physical models and Bayesian learning to accelerate chemical research

Abstract

Current methods of finding optimal experimental conditions, Edisonian systematic searches, often inefficiently evaluate suboptimal design points and require fine resolution to identify near optimal conditions. For expensive experimental campaigns or...

Bayesian optimization of dynamic Ce(III) electrooxidation

“…The BO method used in this study was based on the BO algorithm reported previously 60 . BO algorithms consist of two main components: a surrogate model (SM) and an acquisition function.…”

“…The BO method used in this study was based on the BO algorithm reported previously. 60 After the new batch was selected, the experiments were performed, and the results were added to the known experimental conditions.…”

Automated optimization under dynamic flow conditions

“…Early examples of this technology focused on self-optimization of single-step reactions using black-box optimization algorithms. [17][18][19][20] Growing interest in flow chemistry research and applications, along with advancements in machine learning, 21,22 have led to the expansion of automated optimization systems. 23 Besides optimizing the continuous variables of complex reactions, 24 discrete optimization approaches have been used to identify the best solvent, 25 base, 26 and catalyst 10,27 for reactions.…”

Bayesian optimization of electrochemical devices for electrons-to-molecules conversions: the case of pulsed CO2 electroreduction

“…A promising data-driven optimization strategy to identify global optima with the minimum amount of experimental input is Bayesian optimization (BO). [7][8][9][10][11][12][13][14] BO methods for reactor optimization rely on a surrogate model to statistically predict the mean and uncertainty of a desired performance metric for any possible combination of operating parameters. These surrogate models are then used to decide what experiments will provide the most information from the reactors and allow the identification of the optimum conditions with the minimum number of experiments.…”