Formula used to upweight the loss of a training instance is slightly different from the original paper

[ec0github]

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]

已收到您的邮件！谢谢！