Disentangled Ontology Embedding for Zero-shot Learning

Knowledge Graph (KG) and its variant of ontology have been widely used for representation, and have shown to be quite effective in augmenting-shot Learning (ZSL). However, existing ZSL methods that utilize KGs all the intrinsic complexity of inter-class relationships represented in. One typical feature is that a class is often related to other classes in semantic aspects. In this paper, we focus on ontologies for ZSL, and propose to learn disentangled ontology embeddings guided by properties to capture and utilize more fine-grained class in different aspects. We also contribute a new ZSL framework DOZSL, which contains two new ZSL solutions based on generative models graph propagation models, respectively, for effectively utilizing the ontology embeddings. Extensive evaluations have been conducted on benchmarks across zero-shot image classification (ZS-IMGC) and zero-shot completion (ZS-KGC). DOZSL often achieves better performance than the-of-the-art, and its components have been verified by ablation studies and studies. Our codes and datasets are available at://github.com/zjukg/DOZSL.