Do you have a plan to develop an adversarial defense integration?

QData / TextAttack

TextAttack 🐙 is a Python framework for adversarial attacks, data augmentation, and model training in NLP https://textattack.readthedocs.io/en/master/

https://textattack.readthedocs.io/en/master/

MIT License

2.83k stars 379 forks source link

Do you have a plan to develop an adversarial defense integration? #687

Open yangheng95 opened 1 year ago

jxmorris12 commented 1 year ago

@yangheng95 what kind of integration are you proposing? We already support adversarial training with all of the built-in attack methods.

yangheng95 commented 1 year ago

Hi, The initial idea is provide a interface for allowing deploying an adversarial example post-defense service, which can be used for benchmarking text defense methods. I already implement it but it is not ready for merge. I will PR when it is ready.

jxmorris12 commented 1 year ago

@yangheng95 great, please feel free to submit a pull request!