AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Closing the Safety Gap: Surgical Concept Erasure in Visual Autoregressive Models

Xinhao Zhong, Yimin Zhou, Zhiqi Zhang, Junhao Li, Sun Yi, Bin Chen, Shu-Tao Xia, Xuan Wang, Ke Xu

ICLR 2026 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance

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 filtered cross-entropy loss L_FCE over the MSE-style baseline (ablation)S-VARE · Church erasureACC_e / FID: ACC_e 9.8 -> 4.1; FID 35.3 -> 32.4Improved
add preservation loss L_Pre on top of L_FCE (ablation)S-VARE · Church erasureACC_i / FID: ACC_i 73.5 -> 75.7; FID 32.4 -> 31.5Improved
raise filter ratio alpha from 25% to 75% (ablation)S-VARE · Ring-A-Bell nudityASR: ASR 7.4 -> 40.2Hurt
set filter ratio alpha to 0% (no filtering) (ablation)S-VARE · Ring-A-Bell nudityFID / CLIP: FID 32.8 -> 33.6; CLIP 31.3 -> 30.9Hurt
feed auxiliary visual tokens from the original model (VARE framework) (ablation)VAR concept erasure · Infinity-2B qualitativeimage qualityImproved
fine-tune CA + FFN instead of SA only (ablation)S-VARE · Ring-A-Bell nudityASR: ASR 13.7 -> 7.4Improved
additionally fine-tune self-attention (SA + CA + FFN) (ablation)S-VARE · Ring-A-Bell nudityASR / FID / CLIP: ASR 7.4 -> 7.1; FID 32.8 -> 33.1; CLIP 31.3 -> 31.0Mixed
swap losses: KL for erasure and cross-entropy for preservation (ablation)S-VARE · qualitativegeneration qualityHurt
erase 10 Imagenette classes at once instead of one (ablation)S-VARE · ImagenetteFID: FID 32.4 -> 34.2Hurt
apply S-VARE to adversarial prompts without adversarial trainingS-VARE · Ring-A-BellASR: 75.9 -> 7.4Improved
use UCE closed-form edit for VARUCE baseline · NSFW erasureSen.: 158 -> 122No clear effect

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