Skip to content
    Flow-alerta.cab

    Few-Shot Learning

    AI fundamentals & modelsLearning with few examples
    Editorially reviewed on: 2 sources
    Your reading view

    The short answer · 11 words

    In short

    Few-Shot Learning aims for good adaptation with very few identified examples.

    Flow-alerta.cab (2026): Few-Shot Learning. https://flow-alerta.cab/en/glossar/few-shot-learning. Editorially reviewed on: 10 October 2026.

    How much detail would you like?

    Practical

    Useful for expensive annotation, but highly dependent on pre-training, examples, and task similarity.

    Hypothetical example: Three checked ticket examples show a language model the desired classification format.

    SMEs can test small pilots with a few tested examples, should perform robust tests against zero and many-shot baselines.

    At a glance

    Core
    The term includes procedures that learn from a few examples of a new class or task. In language models, prompting with few demonstrations is often referred to as such, without a change in weight.
    Use
    Useful for expensive annotation, but highly dependent on pre-training, examples, and task similarity.
    Example
    Hypothetical example: Three checked ticket examples show a language model the desired classification format.
    Limits
    Mixing prompting and model training; not choosing examples representatively; ignoring variance over example order.

    In context

    In-Context Learning

    In-context learning means that a model uses examples within the current input.

    Fine-Tuning

    Fine tuning adapts an existing model with additional sample data.

    Questions about this term

    Are Model Weights Changing?

    No at prompting, possible at training.

    Are three examples always enough?

    No.

    All details and background

    The term includes procedures that learn from a few examples of a new class or task. In language models, prompting with few demonstrations is often referred to as such, without a change in weight.

    ELI5 – Explained simply

    A model gets only a few sample examples and should transfer the principle.

    Why it matters in practice

    Useful for expensive annotation, but highly dependent on pre-training, examples, and task similarity.

    Example

    Hypothetical example: Three checked ticket examples show a language model the desired classification format.

    Pitfalls and limits

    • Mixing prompting and model training; not choosing examples representatively; ignoring variance over example order.

    Origin & history

    Snell et al. described 2017 Prototypical Networks for new classes with few examples. Brown et al. demonstrated 2020 Few-Shot Prompting with GPT-3 without gradient updates.

    Use in mid-sized companies

    • SMEs can test small pilots with a few tested examples, should perform robust tests against zero and many-shot baselines.

    Related terms

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI consulting: Plan suitable examples and comparative tests for a small classification pilot.

    Matching tool / use case

    View use casesUse Case: Classification pilot with few examples and variant comparison

    Sources

    Quellenprüfung: 10.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases