Deep transductive network for generalized zero shot learning

Zhang, Haofeng; Liu, Li; Long, Yang; Zhang, Zheng; Shao, Ling

Abstract

Zero Shot Learning (ZSL) aims to learn projective functions on labeled seen data and transfer the learned functions to unseen classes by discovering their relationship with semantic embeddings. However, the mapping process often suffers from the domain shift problem caused by only using the labeled seen data. In this paper, we propose a novel explainable Deep Transductive Network (DTN) for the task of Generalized ZSL (GZSL) by training on both labeled seen data and unlabeled unseen data, with subsequent testing on both seen classes and unseen classes. The proposed network exploits a KL Divergence constraint to iteratively refine the probability of classifying unlabeled instances by learning from their high confidence assignments with the assistance of an auxiliary target distribution. Besides, to avoid the meaningless ascription assumption of unseen data on GZSL, we also propose an experimental paradigm by splitting the unseen data into two equivalent parts for training and testing respectively. Extensive experiments and detailed analysis demonstrate that our DTN can efficiently handle the problems and achieve the state-of-the-art performance on four popular datasets.

Citation

Zhang, H., Liu, L., Long, Y., Zhang, Z., & Shao, L. (2020). Deep transductive network for generalized zero shot learning. Pattern Recognition, 105, Article 107370. https://doi.org/10.1016/j.patcog.2020.107370

• Journal Article Type: Article
• Acceptance Date: Apr 10, 2020
• Online Publication Date: Apr 16, 2020
• Publication Date: Sep 30, 2020
• Deposit Date: Apr 30, 2020
• Publicly Available Date: Apr 16, 2021
• Journal: Pattern Recognition
• Print ISSN: 0031-3203
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 105
• Article Number: 107370
• DOI: https://doi.org/10.1016/j.patcog.2020.107370
• Public URL: https://durham-repository.worktribe.com/output/1303000