AI Termcirca 2008· Added May 30, 2026
Zero-Shot Classification
Zero-shot classification refers to the ability of a model to classify data into categories it was not explicitly trained on.
Zero-shot classification enables AI models to recognize and categorize inputs into classes they were not specifically trained to identify. By leveraging semantic similarities or relationships between known and unknown classes, such systems can extend their knowledge without requiring new labeled datasets. This approach is pivotal in situations where gathering labeled data for every possible category is impractical or impossible.
Examples
- Using GPT-3's capabilities to answer questions about unseen topics by drawing parallels with known subjects.
- Recognizing dog breeds in images without direct examples, using descriptions of their features.
Common misconceptions
- It doesn't mean the model performs perfectly without any reference; semantic understanding is critical.
- It's not simply guessing; zero-shot relies on learned relationships and knowledge transfer.
Related terms
Want more like this?
Open the full library
Fresh AI mastery content every 2 hours.