Research map / Adversarial robustness and security / Certified and empirical robustness
BulletTrain: Accelerating Robust Neural Network Training via Boundary Example Mining
Weizhe Hua, Yichi Zhang, Chuan Guo, Zhiru Zhang, G. Edward Suh
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 |
|---|---|---|---|
| apply BulletTrain to AugMix (JSD loss on 35% of clean examples) | AugMix · CIFAR-10-C | robust accuracy: 88.8% -> 88.8% | No clear effect |
| apply BulletTrain to AugMix | AugMix · CIFAR-100-C | robust accuracy: 64.0% -> 63.6% | No clear effect |
| remove JSD loss for all examples instead of selective BulletTrain (ablation) | AugMix · CIFAR-10-C | robust accuracy: 88.8% -> 86.9% | Hurt |
| remove JSD loss for all examples (ablation) | AugMix · CIFAR-100-C | robust accuracy: 64.0% -> 60.2% | Hurt |
| apply BulletTrain N_R=2 to TRADES lambda=1/6 | TRADES · CIFAR-10 PGD-20 | robust accuracy: 56.61% -> 56.35% | No clear effect |
| apply BulletTrain N_R=2 to TRADES lambda=1/6 | TRADES · CIFAR-10 | clean accuracy: 84.92% -> 85.89% | Improved |
| reduce N_R from 2 to 0 in BulletTrain instead of lowering lambda (ablation) | TRADES · CIFAR-10 PGD-20 | robust accuracy: 56.35% -> 52.10% | Hurt |
| apply BulletTrain N_R=2 to MART | MART · CIFAR-10 PGD-20 | robust accuracy: 57.39% -> 58.74% | Improved |
| apply BulletTrain N_R=1 to MART | MART · CIFAR-10 PGD-20 | robust accuracy: 57.39% -> 58.11% | Improved |
| replace multi-step PGD with YOPO-2-5 in TRADES | TRADES · CIFAR-10 PGD-20 | robust accuracy: 45.28% vs 49.14% | Hurt |
| use single-step FAST training with TRADES | TRADES · CIFAR-10 PGD-20 | robust accuracy: 56.61% -> 4.48% | Hurt |
| use single-step FAST training with MART | MART · CIFAR-10 PGD-20 | robust accuracy: 57.39% -> 0.20% | Hurt |
Closest papers by meaning
- Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness (ICLR 2023 · similarity 0.85)
- Robust Training of Neural Networks against Bias Field Perturbations (AAAI 2023 · similarity 0.85)
- CTBench: A Library and Benchmark for Certified Training (ICML 2025 · similarity 0.84)
- Boundary thickness and robustness in learning models (NeurIPS 2020 · similarity 0.84)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.84)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.83)
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.
