Enhanced Representation-Based Sampling for the Efficient Generation of Data Sets for Machine-Learned Interatomic Potentials

trajectories simulated with uncertainty-driven dynamics and ERBS and find that the latter significantly increases the exploration of configurational space.