Research map / Adversarial robustness and security / Certified and empirical robustness
Boosting Adversarial Training with Hypersphere Embedding
Tianyu Pang, Xiao Yang, Yinpeng Dong, Kun Xu, Jun Zhu, Hang Su
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 hypersphere embedding (FN+WN+angular margin) | PGD-AT · CIFAR-10 | PGD-20 accuracy: 53.97 -> 59.36 | Improved |
| add hypersphere embedding | TRADES · CIFAR-10 | PGD-20 accuracy: 56.48 -> 62.02 | Improved |
| add hypersphere embedding | PGD-AT · CIFAR-10 | clean accuracy: 86.75 -> 86.19 | No clear effect |
| add hypersphere embedding | ALP · CIFAR-10 | feature attack accuracy: 49.55 -> 48.64 | Hurt |
| add hypersphere embedding | FreeAT · ImageNet-C | mCA: 28.22 -> 30.04 | Improved |
| add hypersphere embedding | PGD-AT · CIFAR-10 | adaptive PGD-20 accuracy: 53.97 -> 55.25 | Improved |
| add hypersphere embedding | AT variants · CIFAR-10 | SPSA/NES black-box accuracy: SPSA-80 ALP 65.14 -> 63.89 | Mixed |
| add weight normalization on top of feature normalization (ablation) | PGD-AT+FN · CIFAR-10 | PGD-20 accuracy: 54.46 -> 58.95 | Improved |
| add feature normalization on top of weight normalization (ablation) | PGD-AT+WN · CIFAR-10 | PGD-20 accuracy: 53.93 -> 58.95 | Improved |
| add angular margin (ablation) | TRADES+FN+WN · CIFAR-10 | PGD-20 accuracy: 60.35 -> 62.02 | Improved |
| increase margin from 0.2 to 0.5 (ablation) | TRADES+HE · CIFAR-10 | PGD-20 accuracy: 62.02 -> 56.27 | Hurt |
| increase scale s from 1 to 15 (ablation) | TRADES+HE · CIFAR-10 | PGD-20 accuracy: 50.33 -> 62.02 | Improved |
Closest papers by meaning
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · similarity 0.87)
- Regional Adversarial Training for Better Robust Generalization (IJCV 2024 · similarity 0.86)
- Adversarial Distributional Training for Robust Deep Learning (NeurIPS 2020 · similarity 0.86)
- Subspace Adversarial Training (CVPR 2022 · similarity 0.86)
- Explaining the role of Intrinsic Dimensionality in Adversarial Training (ICML 2025 · similarity 0.85)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.84)
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.
