AI Paper Atlas

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

Membership Inference Attacks Against Fine-tuned Diffusion Language Models

Yuetian Chen, Kaiyuan Zhang, Yuntao Du, Edoardo Stoppa, Charles Fleming, Ashish Kundu, Bruno Ribeiro, Ninghui Li

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Membership inference and inversion · 3 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
use SAMA sign-based subset aggregation instead of loss/ratio attacksLLaDA-8B-Base fine-tuned · MIMIR ArXivAUC: 0.597 -> 0.850Improved
use SAMA instead of RatioLLaDA-8B-Base fine-tuned · MIMIR GitHubTPR@10%FPR: 0.355 -> 0.647Improved
use SAMA instead of Ratio on HackerNewsLLaDA-8B-Base fine-tuned · MIMIR HackerNewsAUC: 0.575 -> 0.657Improved
apply autoregressive MIAs (Loss, Min-K%) to DLMsLLaDA-8B-Base fine-tuned · MIMIRAUC: 0.506 (Loss, ArXiv)No clear effect
use diffusion-image MIAs SecMI/PIALLaDA-8B-Base fine-tuned · MIMIRAUC: 0.520 / 0.525 (ArXiv)No clear effect
add reference model calibration to loss attack (ablation)SAMA · MIMIRAUC: +0.09 to +0.19 AUCImproved
add progressive masking (ablation)SAMA · MIMIRAUC: 2-3% gainsImproved
add robust subset sign-based aggregation (ablation)SAMA · MIMIRAUC: 20-30% AUC increaseImproved
add inverse-step adaptive weighting (ablation)SAMA · MIMIRAUC: 3-5% refinementImproved

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