AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Learning From Dictionary: Enhancing Robustness of Machine-Generated Text Detection in Zero-Shot Language via Adversarial Training

Yuanfan Li, Qi Zhou, Zexuan Xie

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial prompts for language models

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add LAAL on top of adversarial training with RPL (ablation)TASTE (mBERT) · M4GT zero-shot accuracyavg accuracy (%): 70.9 -> 76.2Improved
add LAAL on top of adversarial training with RPL (ablation)TASTE (mBERT) · M4GT under attacksmean ASR (%): 20.8 -> 18.0Improved
add adversarial training (RPL) to mBERT (ablation)mBERT · M4GT under attacksmean ASR (%): 35.5 -> 20.8Improved
add LAAL to adversarial training, German (ablation)TASTE · M4GT Germanaccuracy (%): 50.3 -> 69.5Improved
use gradient-based importance instead of attention-based (ablation)TASTE · M4GT under attacksmean ASR (%): 38.0 -> 18.0Improved
use gradient-based importance instead of random positions (ablation)TASTE · M4GT under attacksmean ASR (%): 30.0 -> 18.0Improved
use gradient-based importance instead of perturbation-based (ablation)TASTE · M4GT under attacksmean ASR (%): 27.6 -> 18.0Improved
use gradient-based importance instead of random positions (ablation)TASTE · M4GT clean accuracymean accuracy: 0.756 -> 0.762No clear effect
increase attack strength ratio from 0.05 to 0.25 (ablation)TASTE · M4GTavg accuracy (%): 76.2 -> 73.1Hurt
increase attack strength ratio from 0.05 to 0.25 (ablation)TASTE · M4GT under attacksavg ASR (%): 18.0 -> 34.2Hurt
introduce 10% wrong dictionary entries (ablation)TASTE · M4GTcross-language mean accuracy (%): 76.2 -> 70.9Hurt
increase dictionary swap rate to 90% (ablation)TASTE · M4GTcross-language mean accuracy (%): 76.2 -> 68.0Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy