AI Paper Atlas

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

Look But Don't Touch with Sparse Autoencoders for Unlearning in Diffusion Models

Enrico Cassano, Riccardo Renzulli, Rayyan Ahmed, Stephan Alaniz, Marco Grangetto

ECCV 2026 · Image generation › Text-to-image diffusion · approach: Multi-concept composition · 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
replace detected patches (PER) instead of latent steeringSAeUron · UnlearnCanvas object unlearningArtifacts Rate (%): 57.0 -> 16.3Improved
replace detected patches (PER) instead of latent steeringSAEmnesia · UnlearnCanvas object unlearningArtifacts Rate (%): 49.8 -> 15.5Improved
replace detected patches (PER) instead of latent steeringG-SAE · UnlearnCanvas object unlearningArtifacts Rate (%): 43.1 -> 22.5Improved
replace detected patches (PER) instead of latent steeringSAE on SDXL Turbo · UnlearnCanvas object unlearningArtifacts Rate (%): 61 -> 8Improved
replace detected patches (PER) instead of latent steeringSAeUron · UnlearnCanvas object unlearningGlobal Average (%): 72.47 -> 84.19Improved
replace detected patches (PER) instead of latent steeringSAEmnesia · UnlearnCanvas object unlearningCross-domain retain accuracy (%): 88.48 -> 97.97Improved
replace detected patches (PER) instead of latent steeringG-SAE · UnlearnCanvas object unlearningUnlearning Accuracy (%): 78.14 -> 94.02Improved
replace detected patches (PER) instead of latent steeringSAeUron · UnlearnCanvas object unlearningUnlearning Accuracy (%): 87.16 -> 85.37Hurt
use padding p=0 instead of p=1 in mask dilation (ablation)SAeUron + PER · UnlearnCanvas object unlearningUnlearning Accuracy (%): 85.37 -> 49.12Hurt
use padding p=2 instead of p=1 (ablation)SAEmnesia + PER · UnlearnCanvas object unlearningAvg of UA/IRA/CRA (%): 93.45 -> 91.70Hurt
use random replacement instead of nearest concept-free patch (ablation)SAEmnesia + PER · UnlearnCanvas object unlearning (p=1)Avg of UA/IRA/CRA (%): 85.07 -> 90.95Improved
apply PER on top of SAEmnesia for NSFW removalSAEmnesia · I2P NSFWexposed body part detections: 47 -> 27Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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