Machine learning potentials (MLPs) have become an indispensable tool for large-scale atomistic simulations, due to their accuracy comparable with ab-initio methods at considerably reduced computational cost. The development ...