AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Trap and Replace: Defending Backdoor Attacks by Trapping Them into an Easy-to-Replace Subnetwork

Haotao Wang, Junyuan Hong, Aston Zhang, Jiayu Zhou, Zhangyang Wang

NeurIPS 2022 · Adversarial robustness and security › Backdoor and poisoning · approach: Training-time backdoor defense · 2 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
remove auxiliary image reconstruction task in stage 1 (ablation)Trap and Replace · CIFAR10 l2-InvisibleASR (%): 0.72 -> 99.99Hurt
remove auxiliary image reconstruction task in stage 1 (ablation)Trap and Replace · CIFAR10 Trojan-WMASR (%): 12.01 -> 51.87Hurt
do not replace the classification head (ablation)Trap and Replace · CIFAR10 l2-InvisibleASR (%): 0.72 -> 99.96Hurt
do not replace the classification head (ablation)Trap and Replace · CIFAR10 SIGASR (%): 0.01 -> 97.24Hurt
stem has 13 layers, head only 1 FC layer (vs 11/3) (ablation)Trap and Replace · CIFAR10 BlendASR (%): 10.59 -> 97.41Hurt
make head larger: 9 stem layers, 5 head layers (vs 11/3) (ablation)Trap and Replace · CIFAR10 l2-Invisibleclean accuracy (%): 84.01 -> 73.50Hurt
make head larger: 9 stem layers, 5 head layers (vs 11/3) (ablation)Trap and Replace · CIFAR10 SmoothASR (%): 4.23 -> 4.17No clear effect
raise reconstruction weight lambda1 from 0 to 10 (ablation)Trap and Replace lambda1 · CIFAR10 l2-InvisibleASR (%): 99.99 -> 0.74Improved
raise lambda1 from 10 to 20 (ablation)Trap and Replace · CIFAR10 l2-Invisibleclean accuracy (%): 84.01 -> 81.37Hurt
set total variation weight lambda2 to 0.1 instead of 0 (ablation)Trap and Replace · CIFAR10 l2-InvisibleASR (%): 3.32 -> 0.74Improved
reduce clean holdout set from 2500 to 250 samplesTrap and Replace · CIFAR10 l2-Invisibleclean accuracy (%): 84.01 -> 72.00Hurt
reduce clean holdout set from 2500 to 1250 samplesTrap and Replace · CIFAR10 l2-Invisibleclean accuracy (%): 84.01 -> 83.31No 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