AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Label Poisoning is All You Need

Rishi Dev Jha, Jonathan Hayase, Sewoong Oh

NeurIPS 2023 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 4 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 flipped labels from 0 to 1500FLIP · CIFAR-10 ResNet-32 sinusoidalpoison test accuracy PTA (%): 00.1 -> 99.8Improved
Increase flipped labels from 0 to 1500FLIP · CIFAR-10 ResNet-32 sinusoidalclean test accuracy CTA (%): 92.38 -> 89.87Hurt
Use pixel trigger instead of sinusoidalFLIP · CIFAR-10 ResNet-32, 1500 labelspoison test accuracy PTA (%): 99.8 -> 29.9Hurt
Use FLIP instead of inner product baselineinner product · CIFAR-10 ResNet-32 sinusoidal, 1500 labelspoison test accuracy PTA (%): 32.0 -> 99.8Improved
Use FLIP instead of random selection baselinerandom selection · CIFAR-10 ResNet-32 sinusoidal, 1500 labelspoison test accuracy PTA (%): 23.0 -> 99.8Improved
Flip 5000 labels with inner product baseline to reach high PTAinner product · CIFAR-10 ResNet-32CTA: 92.38 -> 82.84Hurt
Mismatch expert ResNet-32 and user ResNet-18 (ablation)FLIP · CIFAR-10 sinusoidal, 1500 labelspoison test accuracy PTA (%): 99.8 -> 98.0No clear effect
Mismatch expert ResNet-32 and user VGG (ablation)FLIP · CIFAR-10 sinusoidal, 1500 labelspoison test accuracy PTA (%): 99.8 -> 63.0Hurt
User trains with Adam instead of SGD (ablation)FLIP r32 · CIFAR-10 sinusoidal, 1500 labelspoison test accuracy PTA (%): 99.8 -> 99.0No clear effect
Mismatch both architecture and optimizer (r32 -> r18) (ablation)FLIP · CIFAR-10 sinusoidal, 1000 labelspoison test accuracy PTA (%): 86.9 -> 99.2Improved
Use traditionally poisoned teacher for distillationstudent model · CIFAR-10poison test accuracy PTA (%): 99.9 -> 0.20No clear effect
Train expert models for K=0 epochs (ablation)FLIP · CIFAR-10 sinusoidalpoison test accuracy PTA (%)Hurt

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