Fig. 10From: Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in moleculesConvergence of the training and validation runs for the AISD HOMO-LUMO data on Summit with different GPU countsBack to article page