Few-shot learning is a machine learning paradigm where models are trained to generalize from only a small number of labeled examples per class—often as few as 1 to 5. It addresses the challenge of learning effectively in low-data regimes, which is common in real-world applications like medical diagnosis, rare language processing, or personalization.

Approaches typically involve meta-learning (learning how to learn), transfer learning, or using pretrained models to extract generalizable representations. Common strategies include:

• Metric-based: Learn similarity functions (e.g., Prototypical Networks, Siamese Networks)
• Optimization-based: Adapt quickly to new tasks with minimal gradient steps (e.g., MAML)
• Generative methods: Use generative models to synthesize new examples.

Few-shot tasks are often benchmarked using N-way K-shot setups (e.g., 5-way 1-shot classification). Models are evaluated based on their ability to adapt to new classes unseen during training.

Few-shot learning reduces annotation cost and enables faster deployment in specialized domains with limited labeled data.