AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Enhancing All-to-X Backdoor Attacks with Optimized Target Class Mapping

Lei Wang, Yulong Tian, Hao Han, Fengyuan Xu

AAAI 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Backdoors in large multimodal models · 1 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 a single-trigger All-to-X attack (multiple target classes) instead of All-to-OneBadNets on CIFAR10 ResNet18 · CIFAR10 vs six defensesASR under defense: ABL: 2.0 -> 79.0 (X=1 vs X=10)Improved
increase number of target classes X in A2X attackA2X backdoor, existing mapping · CIFAR10 low poisoning rate 0.2%ASR (%): 87.8 (A2O) -> 50.7 (X=5) -> 43.8 (X=10)Hurt
replace random or cyclic target mapping with similarity-based grouping plus distance-aware target assignmentA2X backdoor attack · CIFAR10, CIFAR100, Tiny-ImageNetASR (%): avg +6.7%, +16.4%, +14.1% on CIFAR10, CIFAR100, Tiny-ImageNetImproved
optimized mapping gains larger with more classesA2X backdoor attack · CIFAR100 X=10 vs CIFAR10ASR (%): avg +25.8% (CIFAR100 X=10); +5.8% (CIFAR10 X=10)Improved
optimized mapping at 0.5% poisoning on Tiny-ImageNetA2X backdoor attack · Tiny-ImageNetASR (%)No clear effect
optimized mapping under six backdoor defensesA2X backdoor attack · CIFAR10 ResNet18 defensesASR (%): better in 58 of 72 settings; gains 0.3% to 43.1%Mixed
use a surrogate model with different architecture or optimizer than victimoptimized A2X attack · CIFAR10 X=5ASR (%): within 3.28% of same-config resultsNo clear effect
keep only one of grouping or target assignment components (ablation)optimized A2X attack · CIFAR10, CIFAR100, Tiny-ImageNetASR (%)Improved

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