In zero-shot learning, at test time the learner is given samples from new classes that were not observed during training, and must correctly predict the new class they belong to. Methods for zero-shot learning usually associate observed and non-observed classes through a form of auxiliary information such as vector embeddings or symbolic attributes for the classes. See Wikipedia - Zero-Shot Learning.

• Larochelle et al 2008 - Zero-data Learning of New Tasks Introduced the idea of “zero-shot” learning.
• Pasupat & Liang 2014 - Zero-shot Entity Extraction from Web Pages (Jeff: I believe this is the first NLP paper to use the term “zero-shot”.)
• Li et al 2018 - A Deep Relevance Model for Zero-Shot Document Filtering The “first deep model to conduct zero-shot document filtering.”

• Few-Shot Learning
• Prompting
• Task Descriptions