3 ms·

You cannot use any prior, let alone literally any regularizer, and say it would work almost just as well. A standard normal prior centered at 0 and one centere

by kgwgk 7y ago

You cannot use any prior, let alone literally any regularizer, and say it would work almost just as well. A standard normal prior centered at 0 and one centered at 42 can give very different results.

mlevental 7y ago
i said almost - that's code for "obviously i'm not talking about pathological regularizers"
- kgwgk 7y ago
  Well, in that case minimizing the (negative) loglikelihood seems principled but you could minimize literally any loss function and it would work almost just as well.
  - mlevental 7y ago
    Lol agreed!