A survey on Zero-shot learning

Fengyi Song, Yi Chen


Zero-shot learning (ZSL) has recently received extensive attention for its potential in achieving scalable object recognition with lower human labor cost relative to traditional supervised learning. However, zero-shot learning is a nontrivial problem, and its feasibility relies on satisfaction of  several important assumptions and conditions, where learning knowledge shareable between seen classes and unseen classes becomes the foundation. A plenty of works are proposed from different views with various formulations while obeying the foundation. We will review the literature of zero-shot learning comprehensively while putting emphasis on analyzing their motivations, assumptions, and exact mechanism for learning transferable knowledge that is helpful for connecting testing images and description of unseen classes. Finally, benchmarks for evaluating and comparing kinds of approaches are discussed mainly involving the datasets, protocols and evaluation measures. We hope this review may shed light on advanced solutions to zero-shot learning.


Zero-shot learning; Knowledge transfer; Survey

Full Text:



  • There are currently no refbacks.

Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 License.

JOS ©: World Science Publisher United States