Does AI for science need another ImageNet Or totally different benchmarks? A case study of machine learning force fields

Open in new window