3 Comments
User's avatar
Joe Corliss's avatar

The importance function is a rescaled logit function, so maybe there is some theoretical foundation behind it.

Avi Chawla's avatar

Yes, Joe, I was also thinking this but wasn't entirely sure about it. But yes, there might be some connection there.

Debayan Biswas's avatar

For a misclassification the new weight should be more than the previous. I think you made a sign error in the reweight calculation formula.