Learning catalytic kinetic models from data: current and emerging methods

Kinetic models are ubiquitous in catalysis and can be formulated using ab initio calculations, kinetic studies, or a combination of experimental and computational sources. This opinion first discusses emerging methods of learning kinetic models from data, such as sparse discovery of governing equations, physics-informed neural networks, residual neural networks, and neural ordinary differential equations, in the context of conventional methods such as rate expressions and microkinetic models. Parameter learning techniques and emerging sources of kinetic data are subsequently presented, culminating in a vision for multimodal closed-loop discovery of kinetic models from spectrokinetic and computational data.