Work

BERTGuard: Robust Text Classification against Adversarial Attacks

Year: 2024

Type: preprint

Abstract: This paper introduces BERTGuard, a text classification framework that enhances the resilience of BERT models against adversarial attacks through feature trimming and sub-sampling within an adversarial training framework. Our approach outperforms the baseline BERT models against various augmented white-box adversarial attacks. Its resiliency is demonstrated using the benchmark IMDB movie review dataset.

DOI: 10.22541/au.172556904.43002725/v1

Authors: Laxmi Shaw, Mohammed Wasim Ansari, Tahir Ekin

Language: English

Topic: Adversarial Robustness in Machine Learning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

OpenAlex ID: W4414899562

Cites:

Cited by:

Related to: