Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Examples Can Be Effective Data Augmentation for Unsupervised Machine Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Dynamically update the constraint coefficient with MinMax rather than fix it and tune by binary search | MINE-based supervised attacks · MNIST | MI: 28.28 -> 51.29 | Helped |
| Dynamically update the constraint coefficient with MinMax rather than fix it and tune by binary search | MINE-based supervised attacks · CIFAR-10 | MI: 13.69 -> 17.14 | Helped |
| Add MINE-UAE augmentation | Sparse Autoencoder · MNIST | reconstruction error: 0.00561 -> 0.00243 | Helped |
| Add L2-UAE augmentation instead of MINE-UAE (ablation) | Dense Autoencoder · MNIST | reconstruction error: 0.00258 -> 0.00286 | Hurt |
| Add MINE-UAE augmentation | Sparse Autoencoder · SVHN | reconstruction error: 0.00887 -> 0.00235 | Helped |
| Replace MINE-UAE with Gaussian augmentation (σ=0.01) (ablation) | Dense Autoencoder · SVHN | reconstruction error: 0.00659 -> 0.00858±0.00232 | Hurt |
| Add MINE-UAE augmentation | Convolutional Autoencoder · SVHN | reconstruction error: 0.00128 -> 0.00095 | Helped |
| Add MINE-UAE augmentation | Adversarial Autoencoder · MNIST | reconstruction error: 0.04785 -> 0.04581 | Helped |
| Add MINE-UAE augmentation | Concrete Autoencoder · MNIST, Fashion MNIST, Isolet, Mice Protein, Activity | accuracy: 94.97% -> 95.41%; 84.92% -> 85.24%; 81.98% -> 82.93%; 89.81% -> 91.2%; 83.38% -> 84.45% | Helped |
| Add MINE-UAE augmentation | Concrete Autoencoder · Coil-20 | accuracy: 98.96% -> 96.88% | Hurt |
| Add MINE-UAE augmentation | SimCLR (ResNet-18) · CIFAR-10 | accuracy improvement: +1.58% | Helped |
| Use the first convolution layer output rather than random sampling for per-sample MINE (ablation) | Per-sample MINE · CIFAR-10 | KID: 14.86 ± 1.45 -> 10.78 | Helped |
Closest papers by meaning
- On Adversarial Training without Perturbing all Examples (ICLR 2024 · similarity 0.89)
- GAT: Generative Adversarial Training for Adversarial Example Detection and Robust Classification (ICLR 2020 · similarity 0.88)
- Constructing Unrestricted Adversarial Examples with Generative Models (NeurIPS 2018 · similarity 0.88)
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.88)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.87)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
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.