Research map / Adversarial robustness and security / Black-box and transfer attacks
Adversarial Defense via Learning to Generate Diverse Attacks
Yunseok Jang, Tianchen Zhao, Seunghoon Hong, Honglak Lee
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 |
|---|---|---|---|
| add recursive generation of perturbations (ablation) | L2L-DA base · MNIST white-box | classification accuracy: +7.85%p | Improved |
| add diversity loss (single-step generation) (ablation) | L2L-DA base · MNIST white-box | Min accuracy: +4.33%p | Improved |
| add diversity loss on top of recursive generation (ablation) | L2L-DA with recursion · MNIST white-box | Min accuracy: 93.97 -> 95.15 (+1.18%p) | Improved |
| add noise input delta to the generator (ablation) | L2L-DA full · MNIST white-box | Min accuracy: 92.46 -> 95.15 (+2.69%p) | Improved |
| add class label y with class-conditional batch norm (ablation) | L2L-DA full · MNIST white-box | Min accuracy: 93.44 -> 95.15 (+1.71%p) | Improved |
| add gradient normalization to generator input (ablation) | L2L-DA full · CIFAR-10 white-box | PGD10 accuracy: about 1% without it | Improved |
| use gradient normalization, class label and noise inputs (L2L-DA base) instead of L2L* (ablation) | generator-based adversarial training · MNIST white-box | Min accuracy: 0.02 -> 86.12 | Improved |
| adversarial training with L2L-DA instead of best baseline | LeNet-5 · MNIST white-box | Min accuracy: +5.68%p | Improved |
| adversarial training with L2L-DA instead of best baseline | ResNet-20 · CIFAR-10 white-box | Min accuracy: +1.60%p | Improved |
| train with L2L-DA instead of PGD10 | LeNet-5 · MNIST black-box | Min accuracy: 94.13 -> 95.08 | Improved |
| adversarial training on PGD10 only | classifier · CIFAR-10 white-box | Min accuracy vs unseen attacks: 37.37 vs 0.32 Min | Hurt |
Closest papers by meaning
- Deep Defense: Training DNNs with Improved Adversarial Robustness (NeurIPS 2018 · similarity 0.93)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.92)
- Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Models (ICLR 2018 · similarity 0.91)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · 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.
