TRAINING DEVICE, METHOD FOR TRAINING, PREDICTION DEVICE, METHOD FOR PREDICTION, AND PROGRAM

摘要

To predict hiding vectors of the whole graph.SOLUTION: The training device for training a network which predicts features of a graph when data of the graph is input includes: a merge part in each layer for merging a first hiding vector of first nodes forming the graph and a second hiding vector of second nodes connected to the first nodes on the basis of first hiding vector, the connection information about the connections of the first nodes, and the second hiding vector; a first network for updating the first vector and the second vector on the basis of the results of the merge; a second network for extracting features of the graph on the basis of the first hiding vector and the second hiding vector output from the first network; a calculation unit for calculating the loss of the features of the graph output from the second network; and a network updating unit for updating at least one of the first network and the second network on the basis of the loss.SELECTED DRAWING: Figure 1