AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Autoregressive Perturbations for Data Poisoning

Pedro Sandoval-Segura, Vasu Singla, Jonas Geiping, Micah Goldblum, Tom Goldstein, David W. Jacobs

NeurIPS 2022 · Adversarial robustness and security › Attacks on text and graphs · approach: Clean-label data poisoning · 8 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
Use sample-wise autoregressive perturbations instead of the strongest listed baseline poisonResNet-18 trained on CIFAR-10 · CIFAR-10test accuracy: 16.84 -> 11.75Improved
Use the same AR processes across datasets instead of the strongest listed baseline poisonResNet-18 poison training · STL-10test accuracy: 17.26 -> 11.65Improved
Use sample-wise AR poisoning instead of the strongest listed baseline poisonResNet-18 poison training · SVHNtest accuracy: 6.39 -> 6.77Hurt
Use AR poisoning instead of the strongest listed baseline poisonResNet-18 poison training · CIFAR-100test accuracy: 2.99 -> 4.24Hurt
Use AR poisoning instead of the strongest listed baseline poisonCIFAR-10 models across architectures · CIFAR-10test accuracy: 16.84 -> 11.75; 9.96 -> 14.90Mixed
Add Cutout, CutMix, or Mixup during trainingResNet-18 trained on AR-poisoned CIFAR-10, epsilon=1 · CIFAR-10test accuracy: 11.75 -> 11.90; 11.75 -> 11.23; 11.75 -> 11.40No clear effect
Reduce AR perturbation size from epsilon=1 to epsilon=0.5ResNet-18 poison training with standard augmentation · CIFAR-10test accuracy: 11.75 -> 14.28Hurt
Increase adversarial-training radius from 0.125 to 0.25ResNet-18 trained on AR-poisoned CIFAR-10 · CIFAR-10test accuracy: $33.22_{\pm0.77}$ -> $57.08_{\pm0.75}$Hurt
Increase adversarial-training radius from 0.25 to 0.50ResNet-18 trained on AR-poisoned CIFAR-10 · CIFAR-10test accuracy: $57.08_{\pm0.75}$ -> $81.27_{\pm2.61}$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