What question did this study set out to answer?

This research aims to develop an efficient method for generating adversarial samples in text using BERT.

January 1, 2020Open Access

BERT-ATTACK: Adversarial Attack Against BERT Using BERT

View Full Paper

Ask AI

Bookmark

This research aims to develop an efficient method for generating adversarial samples in text using BERT.
Proposes BERT-Attack using pre-trained masked language models, specifically BERT.
Validates effectiveness by testing on fine-tuned models across various downstream tasks.
Compares performance against existing state-of-the-art adversarial attack strategies.
BERT-Attack outperforms current methods in success rate and perturbation percentage.
Generated adversarial samples maintain semantic fluency and consistency.
Low computational cost allows for large-scale sample generation.

No takes yet. Share an insight, caveat, or question.

Randomized trial reveals an effective method to generate adversarial text samples, highlighting its potential in misguiding AI models.

Li et al. (2020) studied this question.

View Full Paper

Ask AI

Bookmark