Research map / Adversarial robustness and security / Certified and empirical robustness
Perturbation-Induced Linearization: Constructing Unlearnable Data with Solely Linear Classifiers
Jinlin Liu, Wei Chen, Xiaojin Zhang
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 |
|---|---|---|---|
| pretrain the linear surrogate before generating perturbations (ablation) | PIL · CIFAR-10 Rotation augmentation | clean test accuracy: 27.99 -> 18.15 | Improved |
| pretrain the linear surrogate before generating perturbations (ablation) | PIL · CIFAR-10 ChannelShuffle augmentation | clean test accuracy: 34.13 -> 10.88 | Improved |
| pretrain the linear surrogate before generating perturbations (ablation) | PIL · CIFAR-10 Cutout augmentation | clean test accuracy: 10.90 -> 14.62 | Hurt |
| use PIL instead of CUDA* under budget-matched 8/255 | CUDA* · CIFAR-10 Basic augmentation | clean test accuracy: 87.45 -> 12.87 | Improved |
| use linear surrogate (PIL) instead of deep surrogate REM for generation time | REM · CIFAR-10 | generation time (s): 54.46k -> 40.53 | Improved |
| increase perturbed fraction from 0.1 to 0.9 | PIL · CIFAR-10 partial perturbation | clean test accuracy: 92.95 -> 76.97 | Hurt |
| train on 90% PIL-perturbed vs clean data | ResNet-18 · CIFAR-10 FGSM 8/255 | accuracy drop: -62.44 -> -72.2 | Hurt |
| apply adversarial training against PIL at eps=2/255 | PIL · CIFAR-10 | clean test accuracy: 88.31 -> 78.15 | Hurt |
| use PIL instead of EM under JPEG50 compression | EM · CIFAR-10 | clean test accuracy: 55.12 -> 50.87 | Improved |
| add gradient orthogonality assumption: clean/perturbed gradients orthogonal | PIL partial perturbation theory · CIFAR-10 | clean loss decrease | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- CUDA* — CUDA: Convolution-Based Unlearnable Datasets (CVPR 2023) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Semantic Perturbations With Normalizing Flows for Improved Generalization (ICCV 2021 · similarity 0.88)
- Learning perturbation sets for robust machine learning (ICLR 2021 · similarity 0.88)
- Purify Unlearnable Examples via Rate-Constrained Variational Autoencoders (ICML 2024 · similarity 0.86)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · similarity 0.85)
- Improving robustness to corruptions with multiplicative weight perturbations (NeurIPS 2024 · similarity 0.85)
- Stable Unlearnable Example: Enhancing the Robustness of Unlearnable Examples via Stable Error-Minimizing Noise (AAAI 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.
