Research map / Adversarial robustness and security / Certified and empirical robustness
Learnable Boundary Guided Adversarial Training
Jiequan Cui, Shu Liu, Liwei Wang, Jiaya Jia
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 extra cross-entropy loss on the robust model (ablation) | LBGAT (alpha=0) · CIFAR-10 (PGD-20) | robust accuracy: 57.55% -> 55.50% | Hurt |
| replace MSE logit loss with KL-divergence (ablation) | LBGAT (alpha=0) · CIFAR-10 (PGD-20) | robust accuracy: 57.55% -> 56.10% | Hurt |
| share convolution weights between natural and robust models with separate batch norm (ablation) | LBGAT (alpha=0) · CIFAR-100 (AutoAttack) | robust accuracy: 27.05% -> 24.02% (alpha=6: 29.34% -> 27.26%) | Hurt |
| share convolution weights with separate batch norm (ablation) | LBGAT (alpha=0) · CIFAR-100 (AutoAttack) | natural accuracy: 70.03% -> 64.89% | Hurt |
| add natural-model boundary guidance (BGAT) to vanilla AT | vanilla AT · CIFAR-100 (PGD-20) | robust accuracy: 27.46% -> 30.20% (natural 60.90% -> 67.72%) | Improved |
| make the natural model learnable (LBGAT) instead of static (BGAT) (ablation) | BGAT · CIFAR-100 (PGD-20) | robust accuracy: 30.20% -> 34.30% (natural 67.72% -> 66.29%) | Mixed |
| add LBGAT guidance to ALP | ALP · CIFAR-100 (PGD-20) | robust accuracy: 28.94% -> 35.25% (natural 59.75% -> 62.67%) | Improved |
| add BGAT guidance to ALP | ALP · CIFAR-100 (PGD-20) | robust accuracy: 28.94% -> 31.27% (natural 59.75% -> 63.46%) | Improved |
| add LBGAT guidance to TRADES (alpha=0 vs TRADES alpha=6) | TRADES · CIFAR-100 (AutoAttack) | natural accuracy: 56.50% -> 70.03% (robust 26.87% -> 27.05%) | Improved |
| add LBGAT guidance to TRADES (alpha=0 vs TRADES alpha=6) | TRADES · CIFAR-10 (AutoAttack) | natural accuracy: 84.92% -> 88.22% (robust 52.64% -> 52.86%) | Improved |
| raise TRADES regularization weight alpha from 0 to 6 (ablation) | LBGAT+TRADES · CIFAR-100 (AutoAttack) | robust accuracy: 27.05% -> 29.34% (natural 70.03% -> 60.43%) | Mixed |
| add LBGAT guidance to TRADES (alpha=0 vs TRADES alpha=6) | TRADES · Tiny ImageNet (PGD-20) | natural accuracy: 38.51% -> 47.80% | Improved |
Closest papers by meaning
- Boundary thickness and robustness in learning models (NeurIPS 2020 · similarity 0.90)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.89)
- Towards Better Understanding of Training Certifiably Robust Models against Adversarial Examples (NeurIPS 2021 · similarity 0.89)
- Failure Cases Are Better Learned But Boundary Says Sorry: Facilitating Smooth Perception Change for Accuracy-Robustness Trade-Off in Adversarial Training (ICCV 2025 · similarity 0.88)
- Adversarial Robustness of Supervised Sparse Coding (NeurIPS 2020 · similarity 0.88)
- On the Scalability of Certified Adversarial Robustness with Generated Data (NeurIPS 2024 · similarity 0.87)
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.
