The new model parameters obtained by upweight the loss of a training instance in the book is:
However, the same formula in the original paper is:
They are very similar except one difference. The original formula is to upweight the select instance by ϵ but the formula in the book is to down-weight all the other instance except the selected one by (1 - ϵ). It seems both have the similar purpose but I am interested to know the reason behind the change.
Regards,
Eric
Ericwst
commented
2 years ago
The new model parameters obtained by upweight the loss of a training instance in the book is:
However, the same formula in the original paper is:
They are very similar except one difference. The original formula is to upweight the select instance by ϵ but the formula in the book is to down-weight all the other instance except the selected one by (1 - ϵ). It seems both have the similar purpose but I am interested to know the reason behind the change. Regards, Eric