What does it mean to have a 'masked' loss function?

Web: https://www.reddit.com/r/deeplearning/comments/vgll4a/what_does_it_mean_to_have_a_masked_loss_function/

I am trying to understand the idea behind having masked loss functions or masking a loss function.

For instance, looking at L1_loss functions and I see something called 'Masked L1 Loss'. Any ideas what these are about? Thanks.

deeplearning function loss mean