Question on usage: How to get the sample weights?

[willanxywc]

Hello, what if I want to get the importance weight of each sample in a tensor?

[angeloskath]

Hi,

Do you mean while training? Do you mean after smoothing? In the simple case the importance is proportional to the loss. If you want to compute the importance of each sample in the dataset it is simply the loss value of each sample.

If you look into the implementations we have developed more importance metrics (not yet in the paper).

What are you trying to achieve exactly?

[angeloskath]

Closing because of inactivity.