Machine Learning Meets Chemical Physics

Methods to generate, analyze and rationalize data based on techniques that can be grouped under the broad label of machine learning are gaining an increasingly important place in all fields of science, and chemical physics is no exception. In combination with the increased availability of high-quality, consistent data from high-throughput experiments and calculations, there is little doubt that this is more than a transitory phase. Machine-learning is becoming one of the essential tools in the computational and experimental investigation of the chemistry and physics of molecules and materials.

The algorithms underlying most of the applications to date are rather general, and universally applicable. Being data-driven, however, they benefit from an understanding of the nature and specificity of the data being used. As it gets more deeply embedded into the very fabric of the discipline, machine learning is starting to incorporate insights from chemical physical concepts, and at the same time to generate new insights that become part of our understanding of chemical physics.

This special topic welcomes submissions of papers from all corners of chemical physics that involve the development and the application to both simulations and experiments of machine-learning techniques. These may include classification, regression, experimental design, generative models, supervised and unsupervised learning. We seek in particular work that emphasizes the interplay between machine learning and chemical physics – be it by incorporating physical principles and chemical intuition into the construction of the model, or by using machine learning to recognize new laws or general design principles – as opposed to papers that use off-the-shelf machine learning schemes to sift through data to solve one particular problem.