Zero-shot learning via a specific rank-controlled semantic autoencoder

作者：

Highlights：

• The proposed LSA model solves the domain shift problem, and considers the low-rank structure of the reconstruction data.

• The proposed SRSA model can avoid simultaneously minimizing the variance of reconstruction data.

• Comprehensive experiments on some datasets demonstrate the effectiveness of the proposed approaches.

摘要

•The proposed LSA model solves the domain shift problem, and considers the low-rank structure of the reconstruction data.•The proposed SRSA model can avoid simultaneously minimizing the variance of reconstruction data.•Comprehensive experiments on some datasets demonstrate the effectiveness of the proposed approaches.

论文关键词：Zero-shot learning,Rank,Domain shift,Autoencoder

论文评审过程：Received 9 December 2020, Revised 22 June 2021, Accepted 6 August 2021, Available online 18 August 2021, Version of Record 25 August 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108237