NP-RepMet icon indicating copy to clipboard operation
NP-RepMet copied to clipboard

Published 20 hours ago • verified publisher icon yang-yk

loss in the code is not the same as in the paper

Open kyrawu327 opened this issue 3 years ago • 1 comments

Hi, thank for the nice work and the idea of restoring negative information is quite inspiring. However, I find that the loss in the code is not the same as in the paper, am I missing something? or the code is further optimized?

Looking forward to your kind reply.

kyrawu327 avatar Aug 27 '21 02:08 kyrawu327

Hi, thank for the nice work and the idea of restoring negative information is quite inspiring. However, I find that the loss in the code is not the same as in the paper, am I missing something? or the code is further optimized?

Looking forward to your kind reply.

Yes. The loss is further optimized with the cosine distance. Thanks！

yang-yk avatar Aug 29 '21 02:08 yang-yk