Single-model uncertainty quantification in neural network potentials does not consistently outperform model ensembles

Open in new window