search

thunlp / TAADpapers

Must-read Papers on Textual Adversarial Attack and Defense
MIT License
1.51k stars 195 forks source link

Add one defense paper #18

Closed boxin-wbx closed 3 years ago

boxin-wbx commented 3 years ago

Add one defense paper from ICLR 2021

yangalan123 commented 3 years ago

Hi, sorry for the late reply, and thanks for your contribution! This is a really nice defense work based on the information bottleneck (IB) framework and gives me much inspiration. Glad to see a new defense direction besides typical adversarial training, certification (which optimizes an error bound on a local convert hull, you can also check it in our list :-) ), and input preprocessing (like robust encoding, perturbation recognizer, and so on).