4 ms·

Not everything is a linear combination of large number of (IID) samples, and thus not everything is gaussian distributed.

• kgwgk 1y ago
  You’re implying that many things are though.
  • jampekka 1y ago
    Yes, and I was explicit about it in another comment in this post.
    • kgwgk 1y ago
      Ok, so we all agree that in most cases the reason to minimize square error is that it’s the appropriate thing to minimize - not that it has an analytical solution, convenience or tradition.