Pith. sign in

REVIEW 2 cited by

Fast Adversarial Training against Textual Adversarial Attacks

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2401.12461 v1 pith:FXLCELF4 submitted 2024-01-23 cs.CL cs.AI

Fast Adversarial Training against Textual Adversarial Attacks

classification cs.CL cs.AI
keywords adversarialtrainingperturbationrobustnesssingle-stepascentattacksdefense
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Many adversarial defense methods have been proposed to enhance the adversarial robustness of natural language processing models. However, most of them introduce additional pre-set linguistic knowledge and assume that the synonym candidates used by attackers are accessible, which is an ideal assumption. We delve into adversarial training in the embedding space and propose a Fast Adversarial Training (FAT) method to improve the model robustness in the synonym-unaware scenario from the perspective of single-step perturbation generation and perturbation initialization. Based on the observation that the adversarial perturbations crafted by single-step and multi-step gradient ascent are similar, FAT uses single-step gradient ascent to craft adversarial examples in the embedding space to expedite the training process. Based on the observation that the perturbations generated on the identical training sample in successive epochs are similar, FAT fully utilizes historical information when initializing the perturbation. Extensive experiments demonstrate that FAT significantly boosts the robustness of BERT models in the synonym-unaware scenario, and outperforms the defense baselines under various attacks with character-level and word-level modifications.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Mitigating Error Amplification in Fast Adversarial Training

    cs.LG 2026-04 unverdicted novelty 6.0

    DDG dynamically adjusts perturbation magnitude and supervision strength in fast adversarial training according to sample confidence at the ground-truth class, mitigating catastrophic overfitting and the robustness-acc...

  2. Unveiling the Backdoor Mechanism Hidden Behind Catastrophic Overfitting in Fast Adversarial Training

    cs.LG 2026-04 unverdicted novelty 5.0

    Catastrophic overfitting in fast adversarial training is reinterpreted as a weak-trigger variant of unlearnable tasks, allowing backdoor-inspired recalibration and outlier suppression to restore robustness.