AI Paper Atlas

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

What Can We Learn from Unlearnable Datasets?

Pedro Sandoval-Segura, Vasu Singla, Jonas Geiping, Micah Goldblum, Tom Goldstein

NeurIPS 2023 · Adversarial robustness and security › Attacks on text and graphs · approach: Defenses against universal perturbations

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
reweight the last layer with DFR on features trained on Adversarial Poisoningpoison-trained ResNet-18 · CIFAR-10max DFR test accuracy (%): 35.97 -> 69.99Improved
DFR on features trained on class-wise Unlearnable Examplespoison-trained ResNet-18 · CIFAR-10max DFR test accuracy (%): 35.97 -> 34.09Hurt
DFR on features trained on class-wise Random Noisepoison-trained ResNet-18 · CIFAR-10max DFR test accuracy (%): 35.97 -> 66.79Improved
use ViT instead of ResNet-18 for DFR on Adversarial PoisoningDFR · CIFAR-10max DFR test accuracy (%): 37.40 -> 68.97Improved
use GoogleNet for DFR on Unlearnable ExamplesDFR · CIFAR-10max DFR test accuracy (%): 48.08 -> 41.08Hurt
apply DFR on SVHN Adversarial Poisoning featuresDFR · SVHNmax DFR test accuracy (%): 32.05 -> 87.06Improved
AR perturbations versus other perturbations for linear separabilitylogistic regression probe · CIFAR-10linear train accuracy (%): 39.58 vs 100Mixed
orthogonal projection against class-wise Unlearnable Examples instead of adversarial trainingOrthogonal Projection · CIFAR-10 ResNet-18test accuracy (%): 85.56 -> 89.98Improved
orthogonal projection instead of adversarial training on OPSOrthogonal Projection · CIFAR-10 ResNet-18test accuracy (%): 11.77 -> 87.94Improved
orthogonal projection instead of adversarial training on Adversarial PoisoningOrthogonal Projection · CIFAR-10 ResNet-18test accuracy (%): 85.32 -> 14.74Hurt
orthogonal projection instead of adversarial training on AROrthogonal Projection · CIFAR-10 ResNet-18test accuracy (%): 84.09 -> 13.03Hurt
orthogonal projection on NTGAOrthogonal Projection · CIFAR-10 ResNet-18test accuracy (%): 40.78 -> 82.21Improved

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