Research map / Adversarial robustness and security / Certified and empirical robustness
AdAUC: End-to-end Adversarial AUC Optimization Against Long-tail Problems
Wenzheng Hou, Qianqian Xu, Zhiyong Yang, Shilong Bao, Yuan He, Qingming Huang
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 |
|---|---|---|---|
| Use AdAUC AT2 rather than the strongest CE-based model for adversarial training | WideResNet-28 · CIFAR-10-LT and CIFAR-100-LT, PGD-5/10/20, C&W, and AA | AUC: CIFAR-10-LT PGD-5: 0.4870 -> 0.5591; CIFAR-100-LT PGD-5: 0.3183 -> 0.4644 | Improved |
| Use AdAUC AT2 rather than the strongest CE-based model for adversarial training | WideResNet-28 · MNIST-LT, PGD-5/10/20, C&W, and AA | AUC: PGD-5: 0.8912 -> 0.9436; AA: 0.8818 -> 0.9311 | Improved |
| Use AdAUC AT2 rather than the strongest CE-based model | WideResNet-28 · CIFAR-10-LT clean and MNIST-LT clean | AUC: CIFAR-10-LT: 0.6833 -> 0.7528; MNIST-LT: 0.9736 -> 0.9852 | Improved |
| Use AdAUC AT2 rather than the strongest CE-based model | WideResNet-28 · CIFAR-100-LT clean | AUC: 0.6382 -> 0.6313 | Hurt |
| Add FOSC-based adaptive control of adversarial-example strength (ablation) | AdAUC AT1 · CIFAR-10-LT and CIFAR-100-LT, including PGD-20 | AUC: CIFAR-10-LT PGD-20: 0.4224 -> 0.5283; CIFAR-100-LT PGD-20: 0.2989 -> 0.4065 | Improved |
| Add FOSC-based adaptive control of adversarial-example strength (ablation) | AdAUC AT1 · MNIST-LT | AUC: PGD-10: 0.9395 -> 0.9347; PGD-20: 0.9382 -> 0.9323 | Mixed |
| Use AUC optimization instead of CE classification loss | natural training on long-tail datasets · CIFAR-10-LT and MNIST-LT clean | AUC: CIFAR-10-LT: 0.7264 -> 0.7885; MNIST-LT: 0.9736 -> 0.9904 | Improved |
Closest papers by meaning
- Long-tailed Adversarial Training with Self-Distillation (ICLR 2025 · similarity 0.89)
- Revisiting Adversarial Training Under Long-Tailed Distributions (CVPR 2024 · similarity 0.88)
- TAET: Two-Stage Adversarial Equalization Training on Long-Tailed Distributions (CVPR 2025 · similarity 0.86)
- Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization (AAAI 2025 · similarity 0.86)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.86)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.86)
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.
