Abstract: Positive and unlabeled (PU) learning aims to train a suitable classifier simply based on a set of positive data and unlabeled data. The state-of-the-art methods usually formulate PU learning ...