Fig. 11: Learning curves of the ML models.
From: Thermodynamics and dielectric response of BaTiO3 by data-driven modeling

a The GAP model for energies and atomic force components, trained on a total of 1200 training structures with 258 randomly selected structures used as the validation set and b the SA-GPR polarization model, trained on a total of 640 structures with 200 validation structures.