A Label Inference algorithm considering vertex importance in semi-supervised learning

摘要

The present invention relates to a label inference method and system considering importance of a vertex in semi-supervised learning. The label inference method in semi-supervised learning to learn a model by using both labeled and unlabeled data comprises: a graph building step of converting input data into a graph; and a label inference step of predicting a label of unlabeled data using the built graph. The label inference step is based on the assumption of smoothness in the semi-supervised learning. By combining the importance of each vertex, the label is inferred.