AI Paper Atlas

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

Optimization and Robustness-Informed Membership Inference Attacks for LLMs

Zichen Song, Qixin Zhang, Ming Li, Yao Shu

AAAI 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 citations (OpenAlex)

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 perturbed-input gradient norms to the feature vector (ablation)OR-MIA · Arxiv (LLaMa-3B)MIA accuracy: 0.779 -> 0.933Improved
add perturbed-input gradient norms to the feature vector (ablation)OR-MIA · HackerNews (LLaMa-6B)MIA accuracy: 0.794 -> 0.979Improved
add perturbed-input gradient norms to the feature vector (ablation)OR-MIA · The Pile (Pythia-70M)MIA accuracy: 0.671 -> 0.957Improved
use gradient-norm OR-MIA instead of output-statistic baselines (LOSS)MIA · The Pile (LLaMa-3B)MIA accuracy: 0.581 -> 0.960Improved
use OR-MIA instead of MoPeMIA · Arxiv (LLaMa-6B)MIA accuracy: 0.624 -> 0.874Improved
increase number of perturbations N up to 10 (ablation)OR-MIA · HackerNewsMIA accuracyImproved
take gradients from middle layers instead of shallow or deep layers (ablation)OR-MIA · Arxiv (3B)MIA accuracyImproved
increase calibration samples to 10-20 per class (ablation)OR-MIA · LLaMa-3B across Arxiv/HackerNews/PileMIA accuracyImproved

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