In this paper, we introduce an auxiliary clustering task to explore the structure of the image data, and judiciously weigh unlabeled data to alleviate the influence of ambiguous data on model training ...