AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Training with More Confidence: Mitigating Injected and Natural Backdoors During Training

Zhenting Wang, Hailun Ding, Juan Zhai, Shiqing Ma

NeurIPS 2022 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 12 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
apply NONE vs undefended trainingBadNets single target, five datasets · Injected Trojansaverage ASR (%): 93.34% -> 1.91%Improved
apply NONE vs undefended trainingnatural Trojans · CIFAR-10 / TrojAIASR: 1.75 times lower ASRImproved
NONE instead of ABLlabel-specific BadNets · GTSRB ResNet18ASR (%): 93.93 -> 0.91Improved
NONE instead of DP-SGDnatural Trojan · CIFAR-10 NiNASR (%): 98.85 -> 34.21Improved
increase trigger size from 3*3 to 20*20 (ablation)NONE · CIFAR-10 ResNet18 BadNetsASR (%): 1.07 -> 99.82Hurt
increase poisoning rate from 0.5% to 20% (ablation)NONE · CIFAR-10 ResNet18 BadNetsASR (%): 2.52 -> 1.39No clear effect
use non-ReLU activations (ELU, Tanhshrink, Softplus) (ablation)NONE · CIFAR-10 ResNet18ASR (%): 1.07 -> 1.46 / 5.11 / 2.07No clear effect
apply NONE to benign trainingImageNet-10 VGG16 · non-attackbenign accuracy (%): 88.84 -> 83.41Hurt
lower learning rate in NONE training (ablation)NONE · CIFAR-10 ResNet18ASRMixed

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