zero-shot
term_id: zero_shot
Category: basic_concepts
Definition
Zero-shot learning enables models to generalize to new categories or tasks for which no labeled training data was provided during the initial training phase. This is typically achieved by leveraging semantic embeddings or textual descriptions that link known concepts to unknown ones. It is particularly powerful in large language models and multimodal systems, allowing for flexible adaptation to novel queries without retraining.
Summary
The ability to perform tasks on unseen classes without prior training examples.
Key Concepts
- Generalization
- Semantic embedding
- No labeled data
- Transfer learning
Use Cases
- Classifying new product categories
- Answering questions on unseen topics
- Cross-lingual text classification