Experiments

Training performance

Training for each neural network is performed on a fixed testset with each implemented neural network potential. We repeat each training 5 times with random parameter initialization.

QM9

NNP	average number of epochs	time @ epoch [min:sec]	RMSE test set [kcal/mol]	reported performance [kcal/mol
ANI2x	282	0:10	1.6 +- 0.2	MAE: 0.322848
SchNet		0:50	1.6 +- 0.3
PaiNN		1:10	2.5 +- 1.2
PhysNet		1:00	1.6 +- 0.1
SAKE
TensorNet
MACE