In this paper, the code for obtaining task prototypes using Laplacian smoothing utilizes features from all nodes, including test and validation nodes. Although only training set features are used when calculating the prototype itself, the message propagation process during Laplacian smoothing requires features from test and validation nodes as well. Additionally, the precondition of Theorem 1 states that nodes used for testing and validation cannot be isolated but must be connected to training nodes. I want to confirm my understanding: if a new test graph with new nodes appears and we need to predict their classes, according to this paper's approach, these nodes must be connected to nodes of known classes to retrain the model before predicting their classes. It seems the method cannot directly predict classes without retraining when encountering completely disconnected new nodes. Could you please clarify this point?
Thank you very much for your time and consideration. I greatly appreciate your innovative work in this area and look forward to your insights on this matter.
Best regards
In this paper, the code for obtaining task prototypes using Laplacian smoothing utilizes features from all nodes, including test and validation nodes. Although only training set features are used when calculating the prototype itself, the message propagation process during Laplacian smoothing requires features from test and validation nodes as well. Additionally, the precondition of Theorem 1 states that nodes used for testing and validation cannot be isolated but must be connected to training nodes. I want to confirm my understanding: if a new test graph with new nodes appears and we need to predict their classes, according to this paper's approach, these nodes must be connected to nodes of known classes to retrain the model before predicting their classes. It seems the method cannot directly predict classes without retraining when encountering completely disconnected new nodes. Could you please clarify this point?
Thank you very much for your time and consideration. I greatly appreciate your innovative work in this area and look forward to your insights on this matter.
Best regards