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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| reweight the last layer with DFR on features trained on Adversarial Poisoning | poison-trained ResNet-18 · CIFAR-10 | max DFR test accuracy (%): 35.97 -> 69.99 | Improved |
| DFR on features trained on class-wise Unlearnable Examples | poison-trained ResNet-18 · CIFAR-10 | max DFR test accuracy (%): 35.97 -> 34.09 | Hurt |
| DFR on features trained on class-wise Random Noise | poison-trained ResNet-18 · CIFAR-10 | max DFR test accuracy (%): 35.97 -> 66.79 | Improved |
| use ViT instead of ResNet-18 for DFR on Adversarial Poisoning | DFR · CIFAR-10 | max DFR test accuracy (%): 37.40 -> 68.97 | Improved |
| use GoogleNet for DFR on Unlearnable Examples | DFR · CIFAR-10 | max DFR test accuracy (%): 48.08 -> 41.08 | Hurt |
| apply DFR on SVHN Adversarial Poisoning features | DFR · SVHN | max DFR test accuracy (%): 32.05 -> 87.06 | Improved |
| AR perturbations versus other perturbations for linear separability | logistic regression probe · CIFAR-10 | linear train accuracy (%): 39.58 vs 100 | Mixed |
| orthogonal projection against class-wise Unlearnable Examples instead of adversarial training | Orthogonal Projection · CIFAR-10 ResNet-18 | test accuracy (%): 85.56 -> 89.98 | Improved |
| orthogonal projection instead of adversarial training on OPS | Orthogonal Projection · CIFAR-10 ResNet-18 | test accuracy (%): 11.77 -> 87.94 | Improved |
| orthogonal projection instead of adversarial training on Adversarial Poisoning | Orthogonal Projection · CIFAR-10 ResNet-18 | test accuracy (%): 85.32 -> 14.74 | Hurt |
| orthogonal projection instead of adversarial training on AR | Orthogonal Projection · CIFAR-10 ResNet-18 | test accuracy (%): 84.09 -> 13.03 | Hurt |
| orthogonal projection on NTGA | Orthogonal Projection · CIFAR-10 ResNet-18 | test accuracy (%): 40.78 -> 82.21 | Improved |
Closest papers by meaning
- Detection and Defense of Unlearnable Examples (AAAI 2024 · similarity 0.87)
- Unlearn and Burn: Adversarial Machine Unlearning Requests Destroy Model Accuracy (ICLR 2025 · similarity 0.86)
- How Far Are We from True Unlearnability? (ICLR 2025 · similarity 0.86)
- Reconstruction Attacks on Machine Unlearning: Simple Models are Vulnerable (NeurIPS 2024 · similarity 0.86)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.85)
- Ungeneralizable Examples (CVPR 2024 · similarity 0.85)
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.
