Research map / Adversarial robustness and security / Black-box and transfer attacks
Gray-box Adversarial Training
B. S. Vivek, Konda Reddy Mopuri, R. Venkatesh Babu
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 with Gray-box Adversarial Training instead of FGSM adversarial training | WideResNet-28-10 · CIFAR-100, eps=8/255 | worst-case accuracy Aw: 44.04 -> 65.43 | Improved |
| train with Gray-box Adversarial Training instead of FGSM adversarial training | WideResNet-28-10 · CIFAR-100, eps=16/255 | worst-case accuracy Aw: 18.58 -> 67.41 | Improved |
| train with Gray-box Adversarial Training instead of FGSM adversarial training | ResNet-18 · CIFAR-10, eps=8/255 | worst-case accuracy Aw: 56.09 -> 79.28 | Improved |
| train with Gray-box Adversarial Training instead of FGSM adversarial training | ResNet-18 · CIFAR-10, eps=16/255 | worst-case accuracy Aw: 22.87 -> 60.81 | Improved |
| train with Gray-box Adversarial Training instead of FGSM adversarial training | ResNet-18 · CIFAR-10, eps=2/255 | worst-case accuracy Aw: 87.08 -> 89.46 | Improved |
| train with Gray-box Adversarial Training instead of ensemble adversarial training (set C) | WideResNet-28-10 · CIFAR-100, eps=16/255 | worst-case accuracy Aw: 15.98 -> 67.41 | Improved |
| train with Gray-box Adversarial Training instead of ensemble adversarial training (set B) | ResNet-18 · CIFAR-10, eps=16/255 | worst-case accuracy Aw: 34.84 -> 60.81 | Improved |
| adversarial training with FGSM instead of normal training | WideResNet-28-10 · CIFAR-100, eps=8/255 | worst-case accuracy Aw: 7.09 -> 44.04 | Improved |
| use ensemble adversarial training instead of plain adversarial training | WideResNet-28-10 · CIFAR-100, eps=8/255 | worst-case accuracy Aw: 44.04 -> 36.8 | Hurt |
| use ensemble adversarial training instead of plain adversarial training | ResNet-18 · CIFAR-10, eps=16/255 | worst-case accuracy Aw: 22.87 -> 31.01 | Improved |
| use FGSM instead of FGSM-LL or FGSM-Rand as the attack (ablation) | normally trained WideResNet-28-10 · CIFAR-100 | accuracy under attack | Improved |
| increase drop-in-loss interval D from 0.2 to 0.4 for saving intermediate models (ablation) | GAT ResNet-18 · CIFAR-10 | robustness plot valley | Hurt |
Closest papers by meaning
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.88)
- Adversarial Machine Learning at Scale (ICLR 2017 · similarity 0.88)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- Adversarial Defense via Learning to Generate Diverse Attacks (ICCV 2019 · similarity 0.88)
- Practical Black-box Attacks on Deep Neural Networks using Efficient Query Mechanisms (ECCV 2018 · similarity 0.88)
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.
