Skip to content
    Flow-alerta.cab

    Continual Learning

    AI fundamentals & modelsContinuous learningLifelong Learning
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 14 words

    In short

    Learning new tasks or data while preserving already acquired skills as much as possible.

    Flow-alerta.cab (2026): Continual Learning. https://flow-alerta.cab/en/glossar/continual-learning. Editorially reviewed on: 09 October 2026.

    How much detail would you like?

    Practical

    Relevant for models that are to adapt to changing products, processes or data distributions.

    Example scenario: Defect detection learns new components and is still tested on older component classes.

    Evaluate new and previous skills separately, version training data and enable a return to stable model statuses.

    At a glance

    Core
    A model is adapted over several learning phases. New updates can affect previous abilities, which is known as catastrophic forgetting. Methods such as replay or regularization are intended to limit this conflict. An ongoing chat is not automatically continuous learning of the model weights.
    Use
    Relevant for models that are to adapt to changing products, processes or data distributions.
    Example
    Example scenario: Defect detection learns new components and is still tested on older component classes.
    Limits
    Only test new tasks; enable untested automatic learning in-service; ignore replay data quality and storage costs.

    In context

    Fine-Tuning

    Fine tuning adapts an existing model with additional sample data.

    Concept Drift

    Change in the relationship between input data and target variable, which can make a previously suitable model less reliable over time.

    Questions about this term

    Is every chat a training?

    No.

    Is forgetting excluded?

    No.

    Do you need permanent online training?

    No.

    All details and background

    A model is adapted over several learning phases. New updates can affect previous abilities, which is known as catastrophic forgetting. Methods such as replay or regularization are intended to limit this conflict. An ongoing chat is not automatically continuous learning of the model weights.

    ELI5 – Explained simply

    You learn something new and try not to forget what you have learned so far.

    Why it matters in practice

    Relevant for models that are to adapt to changing products, processes or data distributions.

    Example

    Example scenario: Defect detection learns new components and is still tested on older component classes.

    Pitfalls and limits

    • Only test new tasks; enable untested automatic learning in-service; ignore replay data quality and storage costs.

    Origin & history

    The research has addressed forgetting neural networks for decades; Elastic Weight Consolidation was prominently featured in 2016/2017.

    Use in mid-sized companies

    • Evaluate new and previous skills separately, version training data and enable a return to stable model statuses.

    Related terms

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI consulting – planning learning and evaluation processes for changing company data.

    Matching tool / use case

    View use casesUse case: step-by-step extension of a quality control with new product classes.

    Sources

    • arxiv.orgKirkpatrick et al., Overcoming catastrophic forgetting in neural networks

    Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases