Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Stable and Efficient Training of Verifiably Robust Neural Networks
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 |
|---|---|---|---|
| Combine IBP forward bounds with CROWN backward bounds during training | Certified robust training · MNIST, ε=0.3 | verified error: 8.47 -> 7.02 | Helped |
| Combine IBP forward bounds with CROWN backward bounds during training | Certified robust training · MNIST, ε=0.4 | verified error: 14.80 -> 12.06 | Helped |
| Use CROWN-IBP rather than IBP training | DM-Large · CIFAR-10, ε=2/255 | verified error: 55.19 -> 46.03 | Helped |
| Use CROWN-IBP rather than IBP training | DM-Large · CIFAR-10, ε=8/255 | verified error: 70.81 -> 66.94 | Helped |
| Use CROWN-IBP rather than IBP training | DM-Large · CIFAR-10, ε=16/255 | verified error: 78.91 -> 76.80 | Helped |
| Keep β=1 throughout rather than decrease β from 1 to 0 (ablation) | CROWN-IBP · CIFAR-10, ε=2/255 | verified error: 46.43 -> 46.03 | Helped |
| Use κ 0→0 rather than κ 1→0.5 (ablation) | CROWN-IBP · CIFAR-10, ε=8/255 | verified error: 69.55 -> 66.94 | Helped |
| Use κ 0→0 rather than κ 1→0.5 (ablation) | CROWN-IBP · CIFAR-10, ε=8/255 | standard error: 45.47 -> 54.02 | Hurt |
| Use the revised CIFAR-10 training schedule (ablation) | IBP baseline, DM-Large · CIFAR-10, ε=8/255 | verified error: 73.52% to 72.68% | Helped |
Closest papers by meaning
- Towards Evaluating and Training Verifiably Robust Neural Networks (CVPR 2021 · similarity 0.96)
- Understanding Certified Training with Interval Bound Propagation (ICLR 2024 · similarity 0.92)
- On the Convergence of Certified Robust Training with Interval Bound Propagation (ICLR 2022 · similarity 0.92)
- SDP-CROWN: Efficient Bound Propagation for Neural Network Verification with Tightness of Semidefinite Programming (ICML 2025 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.88)
- Expressive Losses for Verified Robustness via Convex Combinations (ICLR 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.