Deep unsupervised learning with consistent inference of latent representations
作者：
Highlights：
•
• An end-to-end unsupervised method is proposed to pre-train deep neural networks.
• The presented model is optimized under the EM algorithm framework.
• A series of variants of the proposed method are obtained.
• The convergence of our approach and its variants is theoretically verified.
摘要
•An end-to-end unsupervised method is proposed to pre-train deep neural networks.•The presented model is optimized under the EM algorithm framework.•A series of variants of the proposed method are obtained.•The convergence of our approach and its variants is theoretically verified.
论文关键词：Deep unsupervised learning,Consistent inference of latent representations
论文评审过程：Received 6 February 2017, Revised 1 October 2017, Accepted 16 October 2017, Available online 17 October 2017, Version of Record 6 February 2018.
论文官网地址：https://doi.org/10.1016/j.patcog.2017.10.022
