Can a machine learn chemistry?

We are leaving a new paradigm of science—data-driven science motivated by the bid data era. As a result, artificial intelligence, particularly machine learning techniques, has proved very useful in learning and predicting complicated scenarios. One of the main applications of machine learning in atomic, molecular, and optical physics, as well as in chemical physics, is predicting the outcome of chemical reactions. However, one of the main obstacles to reaching that goal is that, in principle, a machine should learn chemistry before making valuable predictions. But how could a machine learn chemistry? In our group, we plan to find an answer to this question by analyzing first the simplest molecules: diatomics. Specifically, we will show that it is possible to predict molecular properties of diatomic molecules (spectroscopic constants and dipole moments) from atomic ones. Apart from delivering very accurate predictions of any diatomic molecule, our research shows that it could be a more efficient way to classify molecules.