• Mandelmus100@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 year ago

    Another big takeaway is that training for more epochs leads to more memorization.

    Should be expected. It’s overfitting.

      • DigThatData@alien.topB
        link
        fedilink
        English
        arrow-up
        1
        ·
        1 year ago

        it’s possible to “overfit” to a subset of the data. generalization error going up is a symptom of “overfitting” to the entire dataset. memorization is functionally equivalent to locally overfitting, i.e. generalization error going up in a specific neighborhood of the data. you can have a global reduction in generalization error while also having neighborhoods where generalization gets worse.

        • Hostilis_@alien.topB
          link
          fedilink
          English
          arrow-up
          1
          ·
          11 months ago

          Memorization is functionally equivalent to locally overfitting.

          Uh, no it is not. Memorization and overfitting are not the same thing. You are certainly capable of memorizing things without degrading your generalization performance (I hope).

        • seraphius@alien.topB
          link
          fedilink
          English
          arrow-up
          1
          ·
          11 months ago

          On most tasks, memorization would be overfitting, but I think one would see that “overfitting” is task/generalization dependent. As long as accurate predictions are being made for new data, it doesn’t matter that it can cough up the old.

    • FaceDeer@alien.topB
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      Indeed. Just like with training humans to be smart, rote memorization sometimes happens but is generally not the goal. Research like this helps avoid it better in future.

    • n_girard@alien.topB
      link
      fedilink
      English
      arrow-up
      1
      ·
      1 year ago

      Hopefully I’m not being offtopic here, but a recent paper suggested that repeating a requirement several times within the same instructions lead the model to be more compliant towards it.

      Do you know whether it’s true or grounded ?

      Thanks in advance.