AI Paper Atlas

Research map / Continual learning / Class-incremental

Can Vision Models Truly Forget? Mirage: Representation-Level Certification of Visual Unlearning

Zhenyu Yu, yangchen zeng, Chunlei Meng, Guangzhen Yao, Shuigeng Zhou

ECCV 2026 · Continual learning › Class-incremental · approach: Class-balanced knowledge retention

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
run BU for more unlearning epochs (1 to 20) (ablation)BU · COVID-19LPR gap vs Retrain: +15.9 -> +15.9 (LPR stays above 94%)No clear effect
run BU for more unlearning epochs (1 to 20) (ablation)BU · CIFAR-10retained accuracy Accr: 79.2 -> 21.8Hurt
run BU for more unlearning epochs (1 to 20) (ablation)BU · CIFAR-10LPR gap vs Retrain: +4.4 -> +0.6Mixed
increase number of passive parties K from 2 to 8 (ablation)BU · CIFAR-10LPR gap vs Retrain: +4.8 -> +3.7No clear effect
increase number of passive parties K from 2 to 8 (ablation)Target · CIFAR-10LPR gap vs Retrain: +1.2 -> -11.9Hurt
increase number of passive parties K from 2 to 8 (ablation)FT · CIFAR-10LPR gap vs Retrain: +7.9 -> +6.2No clear effect
apply BU instead of retrainingBU · COVID-19LPR: 79.2 -> 94.7Hurt
apply FT or SSD unlearningSSD · COVID-19feature separability: .031 -> .340Hurt
apply Target unlearningTarget · COVID-19LPR: 79.2 -> 69.9Mixed
forget class-level vs sample-levelall methods · 7 datasetsLPR: 99% on MNIST, above 96% on COVID-19 vs about 50%Mixed

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