Abstract: This paper addresses the problem of blind distillation, which aims to train a student model with unlabeled data under the supervision of a pre-trained teacher model. The proposed framework ...