Attribution of performance differences to dataset versus architecture
Ascertain whether observed performance differences among neural network potentials arise primarily from differences in training datasets or from architectural choices, by conducting controlled comparisons that isolate the effects of dataset composition from model design.
References
Consequently, it is unclear whether observed differences can be attributed to the training set or the model architecture.
A controlled retraining of competing architectures on our common dataset and CCSD(T) reference would extend this to a full head-to-head accuracy benchmark and is left to future work.
Because several aspects of the two constructions change simultaneously, their individual effects cannot be separated without dedicated ablation studies. In particular, the present comparison does not establish whether the observed size dependence arises from the loss function, training-set composition, descriptors, network architecture, or their combination.