One-Shot Learning

One-shot learning aims to recognize a new category or perform a task from a single example. Rather than learning every concept from many labeled samples, the model relies on representations and prior knowledge acquired during earlier training. A face-recognition system, for example, may compare one enrolled image with future observations, while a language model can infer a requested format from one demonstration in a prompt. Success depends on how representative the example is and how well the pretrained model captures relevant similarities. A single noisy or misleading example can produce fragile behavior, so systems use confidence thresholds, verification, additional context, and evaluation across variations before relying on one-shot decisions.