search

DSE-MSU / DeepRobust

A pytorch adversarial library for attack and defense methods on images and graphs
MIT License
995 stars 192 forks source link

Accelerated DICE #88

Closed sqrhussain closed 3 years ago

sqrhussain commented 3 years ago

By adding edges in batches, we avoid code interpretation overhead

ChandlerBang commented 3 years ago

Hi,

Thanks for the contribution! I just skimmed it and the code seems correct. But before I merge this pull request, I just want to make sure that you have tested that the new DICE resulted in the same graph as the original DICE.

sqrhussain commented 3 years ago

Hi,

I've just tested it, but the resulting graphs were different. The sampling process is completely different but it achieves the same goal. However, if you would like to guarantee backwards compatibility, I'd suggest you don't merge this pull request :)

ChandlerBang commented 3 years ago

This is a neat implementation and I have merged the pull request. Thanks!