Machine learning (ML) has emerged as an invaluable approach for deriving predictive models in the catalysis field. While they are successful in making accurate predictions, many ML models are complex and difficult to interpret. In this opinion, we discuss recent progress in the development of explainable ML models in catalysis. In particular, we focus on the prospect of using symbolic regression (SR) to derive physical models that are based on analytical functional forms rooted in fundamental physics. We overview the basic concepts underlying two popular SR methods (genetic algorithms and compressed sensing), as well as provide recent examples of their application in the catalysis literature.

机器学习 (ML) 已成为在催化领域推导预测模型的宝贵方法。尽管它们成功地做出了准确的预测，但许多 ML 模型很复杂且难以解释。在这个观点中，我们讨论了催化中可解释的ML 模型开发的最新进展。特别是，我们专注于使用符号回归 (SR) 来推导基于根植于基础物理学的分析函数形式的物理模型的前景。我们概述了两种流行的 SR 方法（遗传算法和压缩传感）的基本概念，并提供了它们在催化文献中应用的最新示例。