Skip to content

    Mixture of Recursion

    AI fundamentals & modelsMixture-of-RecursionsMoR
    Reading time: 2 min.Editorially reviewed on:
    ← All terms

    Definition

    A model can distribute computational depth differently per token.

    The research approach combines shared parameters with dynamically chosen recursive processing. Simpler and harder tokens receive different computation depths.

    Simply explained

    A text segment is sent through the same processing station only as many times as the router specifies.

    Why it matters in practice

    This investigates more efficient use of parameters and computational power.

    Example from business

    Hypothetical: A research team compares a recursive model with a standard model of equal cost.

    Limits and pitfalls

    • Research results cannot be directly transferred to arbitrary production models.

    Origin & history

    The approach was published in 2025 as Mixture-of-Recursions.

    Use in mid-sized companies

    • It is especially relevant for SMEs when classifying new model architectures.

    Frequently asked questions

    Is this a prompt technique?

    No, it concerns the architecture and execution of the model.

    Sources

    Editorially reviewed on:

    On this page

    Next step

    Put knowledge into practice