3 ms·i said almost - that's code for "obviously i'm not talking about pathological regularizers"by mlevental 7y agoi said almost - that's code for "obviously i'm not talking about pathological regularizers"kgwgk 7y agoWell, 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 agoLol agreed!