Attacks on NLP

Gradient Masking

Defense technique that modifies the model's gradient to prevent optimization-based attacks, without necessarily improving actual robustness.

← Indietro