Research map / Adversarial robustness and security / Certified and empirical robustness
Attribute-Guided Adversarial Training for Robustness to Natural Perturbations
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 |
|---|---|---|---|
| train on attribute-guided adversarial perturbations instead of pixel-level domain augmentation | CLEVR-Singles color classification · CLEVR-Singles Mat.+Pos. | accuracy: 65.50 -> 69.49 | Helped |
| train with attribute-guided perturbations rather than pixel-level perturbation sets | MNIST classifier · MNIST-RTS T | accuracy: 45.36 -> 52.95 | Helped |
| train with attribute-guided perturbations rather than pixel-level domain augmentation | CIFAR-10 ResNet-26 · CIFAR-10-C Avg. | accuracy: 70.5 -> 72.3 | Helped |
| use blur and noise together rather than blur only (ablation) | AGAT on CIFAR-10 · CIFAR-10-C Avg. | accuracy: 70.0 -> 72.3 | Helped |
| use blur and noise together rather than noise only (ablation) | AGAT on CIFAR-10 · CIFAR-10-C Avg. | accuracy: 71.3 -> 72.3 | Helped |
| add consistency regularization to the classification loss (ablation) | AGAT on MNIST · MNIST-RTS RTS | accuracy: 37.80 -> 41.43 | Helped |
| increase augmented samples from 10% to 70% (ablation) | AGAT on MNIST with N_aug 1 · MNIST-RTS RTS | accuracy: 34.12 -> 51.13 | Helped |
| increase augmented samples from 10% to 70% (ablation) | AGAT on MNIST with N_aug 2 · MNIST-RTS RTS | accuracy: 36.84 -> 50.28 | Helped |
| use AGAT with blur only rather than blur and noise (ablation) | AGAT on CIFAR-10 · CIFAR-10-C N | accuracy: 67.7 -> 47.4 | Hurt |
| use AGAT with noise only rather than blur and noise (ablation) | AGAT on CIFAR-10 · CIFAR-10-C W | accuracy: 79.0 -> 76.1 | Hurt |
| use AGAT rather than MADA | MNIST classifier · MNIST-RTS S | accuracy: 98.32 -> 96.11 | Hurt |
Closest papers by meaning
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.92)
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.91)
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.91)
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.90)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.90)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.90)
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.