Machine learning of chemical transformations in the Si-graphene system from atomically resolved images via variational autoencoder

We suggest and implement an approach for the bottom-up description of systems undergoing large-scale structural changes and chemical transformations from dynamic atomically resolved imaging data, where only partial or uncertain data on atomic positions are available. This approach is predicated on the synergy of two concepts, the parsimony of physical descriptors and general rotational invariance of non-crystalline solids. The first concept is implemented here via the classical variational autoencoder applied to semantically segmented atom-resolved data, seeking the most effective reduced representation for the system that still contains the maximum amount of original information. The second concept is implemented using the rotationally invariant extension of the variational autoencoder. This approach allowed us to explore the dynamic evolution of electron beam induced processes in a silicon-doped graphene system, but it can be also applied for a much broader range of atomic-scale and mesoscopic phenomena. The interactive Jupyter notebook that goes through the analysis described in the paper is available at this https URL.