Equipping data-driven experiment planning for Self-driving Laboratories with semantic memory: case studies of transfer learning in chemical reaction optimization

Abstract

SeMOpt uses meta-/few-shot learning to enable knowledge transfer from previous experiments to accelerate Bayesian optimization of chemical reactions.

Artificial Intelligence (AI) Workflow for Catalyst Design and Optimization

“…Moreover, GP predominantly operates on continuous variables, necessitating the conversion of categorical variables into a continuous format. , During the initial stages of optimization catalyst synthesis, the volume of experimental data is typically limited, especially in the field of the development of new catalysts. In these low-data regimes, simpler representation methods such as one-hot encodings (OHE) often yield significant results, or even surpass, the performance of more complex and resource-intensive descriptors. , The handling of categorical variables which are identified by the first step of the AI workflow as some of the important variables, using the method of one-hot encoding, adds to the flexibility of the GP model, broadening its applicability to a wider range of scenarios. For example, in the synthesis of catalysts, where various solvents and metals are used, these categorical variables can be suitably represented in a binary form via one-hot encoding.…”

“…In these low-data regimes, simpler representation methods such as one-hot encodings (OHE) often yield significant results, or even surpass, the performance of more complex and resource-intensive descriptors. 38,39 The handling of categorical variables which are identified by the first step of the AI workflow as some of the important variables, using the method of one-hot encoding, adds to the flexibility of the GP model, broadening its applicability to a wider range of scenarios. For example, in the synthesis of catalysts, where various solvents and metals are used, these categorical variables can be suitably represented in a binary form via one-hot encoding.…”

Substrate specific closed-loop optimization of carbohydrate protective group chemistry using Bayesian optimization and transfer learning

“…In recent years, some examples of using transfer learning approaches to improve Bayesian optimization have been described. 46,47 However, to the best of our knowledge, transfer learning has not been applied to Bayesian optimization of chemical reactions. Though, transfer learning has been shown to successfully accelerate reaction optimization performed by random forest classiers 48 and neural processes combined with Bayesian optimization.…”

Bayesian optimisation for additive screening and yield improvements – beyond one-hot encoding

“…[18][19][20][21][22][23][24][25][26] Recently, research has focused on sequential model-based optimisation algorithms, particularly Bayesian optimisation (BO), to identify optimal conditions for chemical reactions effectively. [27][28][29][30][31][32][33][34][35][36][37][38] As demonstrated in the space of chemical reactions, BO is particularly well suited for trading off exploration and exploitation in the low data regime. Surprisingly, most BO studies report one-hot encoding (OHE), that contains limited chemical information, to perform remarkably well.…”

“…Surprisingly, most BO studies report one-hot encoding (OHE), that contains limited chemical information, to perform remarkably well. 31,35 This recurring observation raises an important question: why does OHE, with its inherent simplicity manage to deliver competitive results? For instance, Shields et al 32 compared OHE to more elaborate reaction representations such as quantum mechanical (QM) descriptors.…”