Research map / Adversarial robustness and security / Certified and empirical robustness
AROID: Improving Adversarial Robustness Through Online Instance-Wise Data Augmentation
Lin Li, Jianing Qiu, Michael Spratling
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 Vulnerability objective to policy learning (ablation) | AROID · CIFAR-10 (PRN18) | AutoAttack robustness: 49.95 -> 50.57 | Improved |
| add Affinity objective to policy learning (ablation) | AROID · CIFAR-10 (PRN18) | AutoAttack robustness: 49.41 -> 50.57 | Improved |
| add Diversity objective to policy learning (ablation) | AROID · CIFAR-10 (PRN18) | AutoAttack robustness: 22.47 -> 50.57 | Improved |
| add Diversity objective to policy learning (ablation) | AROID · CIFAR-10 (PRN18) | clean accuracy: 73.88 -> 84.68 | Improved |
| learn the augmentation policy instead of uniform sampling from the same space (ablation) | AROID · CIFAR-10 (PRN18) | AutoAttack robustness: 49.18 -> 50.57 | Improved |
| replace RandomCrop with AROID augmentation | vanilla AT · CIFAR-10 (WRN34-10) | AutoAttack robustness: 52.26 -> 55.91 | Improved |
| replace RandomCrop with AROID augmentation | vanilla AT · CIFAR-10 (ViT-B/4) | AutoAttack robustness: 46.72 -> 51.25 | Improved |
| replace RandomCrop with AROID augmentation | vanilla AT · ImageNet (ConvNeXt-T) | robust accuracy: 36.22 -> 40.40 | Improved |
| add AWP and SWA to AROID | AROID · CIFAR-10 (WRN34-10) | AutoAttack robustness: 55.91 -> 57.03 | Improved |
| use offline transferred policy (AROID-T) instead of online AROID | AROID · CIFAR-10 (WRN34-10) | AutoAttack robustness: 55.91 -> 55.61 | No clear effect |
| reduce policy update frequency, K from 5 to 20 (ablation) | AROID · CIFAR-10 (PRN18) | AutoAttack robustness: 50.57 -> 50.45 | No clear effect |
| enlarge the target model to WRN34-12 instead of using AROID | RandomCrop AT · CIFAR-10 | AutoAttack robustness: 52.26 -> 52.46 | No clear effect |
Closest papers by meaning
- Consistency Regularization for Adversarial Robustness (AAAI 2022 · similarity 0.92)
- Data Augmentation Can Improve Robustness (NeurIPS 2021 · similarity 0.91)
- Data augmentation alone can improve adversarial training (ICLR 2023 · similarity 0.91)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Revisiting Adversarial Training at Scale (CVPR 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.
