Differentially Private Instance Encoding against Privacy Attacks
Shangyu Xie, Yuan Hong · 2022
TextHide was proposed to protect the training data via instance encoding in the natural language domain.Due to the lack of theoretic privacy guarantee, such instance encoding scheme has been shown to be vulnerable to privacy attacks, e.g., reconstruction attacks.To address such limitation, we integrate differential privacy into the instance encoding scheme, and thus provide a provable guarantee against privacy attacks.The experimental results also show that the proposed scheme can defend against privacy attacks while ensuring learning utility (as a trade-off).