AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases

Ren Wang, Gaoyuan Zhang, Sijia Liu, Pin-Yu Chen, Jinjun Xiong, Meng Wang

ECCV 2020 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 117 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
increase validation data per class from 1 to 5 (ablation)DL-TND · CIFAR-10AUC: 0.96, 0.98 and 1Improved
increase data poisoning ratio from 0.5% to 10% (higher attack success) (ablation)DL-TND · poisoning ratio sweepAUC: 0.82 -> 0.99Improved
increase data poisoning ratio from 0.5% to 10% (higher attack success) (ablation)DF-TND · poisoning ratio sweepAUC: 0.7 -> 0.99Improved
use DL-TND instead of Neural Cleansedetection of clean networks · Total of 85 models (T1 = 0.7)correctly detected models: 54/85 -> 84/85Improved
use DL-TND instead of Neural Cleansedetection of Trojan networks · Total of 85 models (T1 = 0.7)correctly detected models: 48/85 -> 82/85Improved
use MAD detection rule instead of fixed threshold T1 = 0.7 (ablation)DL-TND · Total of 85 models, clean networkscorrectly detected models: 84/85 -> 67/85Hurt
use MAD detection rule instead of fixed threshold T1 = 0.7 (ablation)DL-TND · Total of 85 models, Trojan networkscorrectly detected models: 82/85 -> 74/85Hurt
use DL-TND instead of NC under MAD ruledetection of Trojan networks · Total of 85 models (MAD rule)correctly detected models: 48/85 -> 74/85Improved
use random noise inputs instead of clean validation data as seeds (ablation)DF-TND · CIFAR-10 modelAUC: 1 -> 0.99No clear effect
use Q = 0.75 instead of Q = 0.25 detection index (ablation)DL-TND · Total, target label detectionAUC: 0.99 -> 0.98Hurt
replace DF-TND by DL-TND at low attack success (30% ASR, 0.5% poisoning)detector choice · poisoning ratio 0.5%AUC: 0.7 -> 0.82Improved
add refine step maximizing the Trojan-related coordinate (ablation)DF-TND · recovered inputstarget label logitsImproved

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