Fig. 1: Training strategy of transferable MLP for hydrogen-carbon system and target-oriented finetuning. | npj Computational Materials

Fig. 1: Training strategy of transferable MLP for hydrogen-carbon system and target-oriented finetuning.

From: Transferable machine learning model for multi-target nanoscale simulations in hydrogen-carbon system from crystal to amorphous

Fig. 1

a Schematic visualization of the training strategy of our transferable MLP for hydrogen-carbon system, which concluded a pre-training process for separated sub-feature spaces and the target-oriented finetuning. b Schematic diagram of convergence process of MLP in structure space during active learning workflow. There are three major sub-feature spaces in this work: a-C, a-C:H and crystal space with the data points represented in different shapes in Fig. 1b, while the secondary sub-feature spaces, e.g., bulk phase and different surface states, are not shown in this schematic diagram.

Back to article page