Research map / Adversarial robustness and security / Certified and empirical robustness
Learning to Confuse: Generating Training Time Adversarial Data with Auto-Encoder
Ji Feng, Qi-Zhi Cai, Zhi-Hua Zhou
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 |
|---|---|---|---|
| Train on bounded adversarial data generated to reduce clean-test performance | CNN · MNIST test | accuracy: $99.32\pm0.05$ -> $0.25\pm0.04$ | Hurt |
| Train on bounded adversarial data generated to reduce clean-test performance | CNN · ImageNet (bulbul vs. jellyfish) test | accuracy: $88.5\pm2.32$ -> $54.2\pm11.19$ | Hurt |
| Train on bounded adversarial data generated to reduce clean-test performance | CNN · CIFAR-10 test | accuracy: $77.28\pm 0.17$ -> $28.77\pm2.80$ | Hurt |
| Use a label-specific objective to shift predictions to predefined target labels | Deep Confuse · MNIST test | targeted attack success rate: 0.00 -> $79.7\pm0.38$ | Improved |
| Use a label-specific objective to shift predictions to predefined target labels | Deep Confuse · MNIST test | accuracy: $0.25\pm0.04$ -> $1.48\pm0.21$ | Hurt |
| Evaluate classifiers trained on generated noise alone | Deep Confuse-trained CNN · MNIST test | accuracy using noise-only inputs: 95.15 | Improved |
| Evaluate classifiers trained on generated noise alone | Deep Confuse-trained CNN · ImageNet test | accuracy using noise-only inputs: 93.00 | Improved |
| Evaluate classifiers trained on generated noise alone | Deep Confuse-trained CNN · CIFAR-10 test | accuracy using noise-only inputs: 72.98 | Improved |
| Transfer generated adversarial training data to classifiers with different architectures or model types | Deep Confuse · MNIST, CIFAR-10, and ImageNet | test performance | Hurt |
Closest papers by meaning
- NAG: Network for Adversary Generation (CVPR 2018 · similarity 0.88)
- Defense Against Universal Adversarial Perturbations (CVPR 2018 · similarity 0.87)
- Generative Adversarial Perturbations (CVPR 2018 · similarity 0.87)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.87)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.86)
- Detecting Adversarial Data Using Perturbation Forgery (CVPR 2025 · 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.
