Fig. 1: Transfer learning pipeline.
From: Transferring chemical and energetic knowledge between molecular systems with machine learning
The top part of the figure represents the training of the neural network model, where the hypergraph representation of the molecules used for training (e.g., examples of the tri-alanine system) are passed through hypergraph message-passing layers to obtain hidden representations. Such representations are further processed by a pooling layer to output the probability of the input being a low free-energy conformation. The bottom part of the figure describes the transfer learning process, where the trained model is used to process examples of the target system (e.g., the deca-alanine system) and make predictions accordingly.
