Deep transductive network for generalized zero shot learning

作者：

Highlights：

• A Deep Transductive Network is proposed to solve the domain shift problem of ZSL.

• KL Divergence is exploited as the soft assignment for unlabeled unseen data with an auxiliary target distribution.

• New experimental paradigm is demonstrated to solve the meaningless assumption for transductive GZSL.

• Extensive experiments show the superiority of the proposed algorithm.

摘要

•A Deep Transductive Network is proposed to solve the domain shift problem of ZSL.•KL Divergence is exploited as the soft assignment for unlabeled unseen data with an auxiliary target distribution.•New experimental paradigm is demonstrated to solve the meaningless assumption for transductive GZSL.•Extensive experiments show the superiority of the proposed algorithm.

论文关键词：Generalized zero shot learning (GZSL),Transductive ZSL,KL Divergence,Deep transductive network (DTN)

论文评审过程：Received 1 June 2019, Revised 1 March 2020, Accepted 10 April 2020, Available online 16 April 2020, Version of Record 5 June 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107370