AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Indelible Backdoors: On the Limits of Post-Training Defenses

Jingyi Guo, Thuy Dung Nguyen, Taylor T Johnson, Kevin Leach

ECCV 2026 · Adversarial robustness and security › Backdoor and poisoning · approach: Training-time backdoor defense · 0 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
align backdoor and clean gradients on Fisher-important neuronstrigger generator backdoor attack · CIFAR-10 under FSTASR: COMBAT 30.65 -> Ours 99.58Improved
use gradient-aligned backdoor against fine-tuning defensesbackdoor attack · CIFAR-10 under FT/NAD/FST/TSBDASR: ASR above 99.5% under all four defensesImproved
pruning defense ANP applied to the aligned backdoorOurs · CIFAR-10 under ANPASR: 100.00 -> 46.50Hurt
increase perturbation budget epsilon from 0.03 to 0.1 (ablation)Ours · CIFAR-10 under TSBDASR: 47.76 -> 99.84Improved
change perturbation budget on clean accuracy (ablation)Ours · CIFAR-10 pretrainedclean accuracy: 92.82 to 93.27 across epsilon 0.075-0.1; 92.91 at 0.03No clear effect
lower poisoning ratio from 0.1 to 0.01 (ablation)Ours · CIFAR-10 under defensesASR: FST 99.58 -> 99.29No clear effect
lower poisoning ratio from 0.1 to 0.01 (ablation)Narcissus · CIFAR-10 pretrainedASR: 94.64 -> 38.26Hurt
reduce defender fine-tuning data ratio (ablation)defenses vs Ours · CIFAR-10ASR: I-BAU 99.63 -> 31.72 at ratio 0.01; FST 99.46 -> 100.00Mixed
evaluate on ViT-B/16 instead of PreAct-ResNet18 (ablation)Ours · CIFAR-10 under FT, TSBDASR: FT 99.99 -> 10.28; TSBD 99.84 -> 14.04Hurt
evaluate on Tiny-ImageNet vs CIFAR-10Ours under FST · Tiny-ImageNetASR: 99.58 -> 21.38 with ACC 32.74Mixed

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