DEAL selects non-redundant structures from atomistic trajectories via Sparse Gaussian Processes (SGP), to be used to train machine-learning interatomic potentials.