Few-Shot Learning Explainer
Few-shot learning describes settings in which a model must adapt to a task or class from only a small number of examples. A support set provides the limited demonstrations, and the system is evaluated on new query examples that were not part of that small set.
Methods include metric learning, which compares a query with learned representations of the support examples; meta-learning, which trains across many small tasks; and in-context learning, where a pretrained language model receives demonstrations in its prompt without updating parameters.
The approach is useful when labels are expensive, rare categories change quickly, or every user has a specialized task. Good prior representations can make limited evidence more useful, but the examples must still show the distinctions that matter.
A handful of examples can be unrepresentative, ambiguous, or sensitive to their order and wording. Few-shot performance should be tested across many sampled support sets, realistic class imbalance, unfamiliar domains, and adversarial or confusing cases—not a single convenient prompt.
Few-shot learning aims to perform a new task or recognize a new class from only a small number of labeled examples.
Key terms include support set, query example, one-shot learning, metric learning, prototype, meta-learning, demonstration and in-context learning.
Metric-based systems place examples in a representation space and classify a query by similarity to support examples or class prototypes.
Meta-learning exposes a model to many training episodes that resemble the few-shot condition so it learns an adaptation strategy rather than one fixed label set.
Performance may change when one demonstration is replaced, reordered, or mislabeled because each example carries unusual influence.
Reliable evaluation repeats trials with different support sets, records uncertainty, includes hard negatives, and checks whether the supposed task was already present in pretraining.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.