Semi-supervised Dual-Branch Network for image classification
作者：
Highlights：
• We present a rarely discussed learned feature distribution mismatch issue between labeled data and unlabeled data.
• We design a dual-branch architecture and a co-consistency regularization to explicitly address mismatch issue.
• We design an augmentation supervised loss for preventing overfitting.
• The results of our method outperform the related works and the state-of-the-arts.
摘要
•We present a rarely discussed learned feature distribution mismatch issue between labeled data and unlabeled data.•We design a dual-branch architecture and a co-consistency regularization to explicitly address mismatch issue.•We design an augmentation supervised loss for preventing overfitting.•The results of our method outperform the related works and the state-of-the-arts.
论文关键词：00-01,99-00,Semi-supervised learning,Deep learning,Learned feature distribution mismatch
论文评审过程：Received 27 September 2019, Revised 14 February 2020, Accepted 29 March 2020, Available online 8 April 2020, Version of Record 24 April 2020.
论文官网地址：https://doi.org/10.1016/j.knosys.2020.105837
