Research map / Adversarial robustness and security / Certified and empirical robustness
LOT: Layer-wise Orthogonal Training on Improving l2 Certified Robustness
Xiaojun Xu, Linyi Li, Bo Li
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 |
|---|---|---|---|
| replace SOC with LOT layer | LipConvnet-20 · CIFAR-10 certified acc rho=108/255 | certified accuracy: 36.04% -> 36.34% | Improved |
| replace SOC with LOT layer at depth 40 | LipConvnet-40 · CIFAR-10 vanilla accuracy | accuracy: 72.59% -> 76.75% | Improved |
| replace SOC with LOT at depth 40, certified rho=36/255 | LipConvnet-40 · CIFAR-10 | certified accuracy: 59.04% -> 63.50% | Improved |
| replace SOC with LOT | LipConvnet-20 · CIFAR-100 vanilla accuracy | accuracy: 47.84% -> 48.81% | Improved |
| LOT pre-computes Newton iteration, replacing SOC exponential | LipConvnet-40 · evaluation time | seconds: 8.668 -> 3.595 | Improved |
| add residual connection to LOT (ablation) | LOT LipConvnet-20 · CIFAR-10 certified acc rho=108/255 | certified accuracy: 30.95% -> 32.12% | Improved |
| add residual connection to SOC (ablation) | SOC LipConvnet-20 · CIFAR-10 certified acc rho=108/255 | certified accuracy: 31.08% -> 30.29% | No clear effect |
| zero padding via pre-processing instead of circular padding (ablation) | LOT LipConvnet-20 · CIFAR-10 certified acc rho=108/255 | certified accuracy: 28.78% -> 32.12% | Improved |
| LOT vs SOC under l_inf PGD, depth 40 | LipConvnet-40 · CIFAR-10 robust accuracy | robust accuracy: 11.68% -> 28.67% | Improved |
| LOT vs SOC under l_inf PGD, depth 5 | LipConvnet-5 · CIFAR-10 robust accuracy | robust accuracy: 27.18% -> 27.19% | No clear effect |
| add 500K unlabelled semi-supervised data, large radius | LOT LipConvnet-20 · CIFAR-10 certified acc rho=108/255 | certified accuracy: 36.34% -> 42.39% | Improved |
| add semi-supervised data, vanilla accuracy | LOT LipConvnet-20 · CIFAR-10 | vanilla accuracy: 77.12% -> 71.86% | Hurt |
Closest papers by meaning
- Improved techniques for deterministic l2 robustness (NeurIPS 2022 · similarity 0.92)
- Improved deterministic l2 robustness on CIFAR-10 and CIFAR-100 (ICLR 2022 · similarity 0.90)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.89)
- Enhancing Certified Robustness via Block Reflector Orthogonal Layers and Logit Annealing Loss (ICML 2025 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.87)
- 1-Lipschitz Layers Compared: Memory Speed and Certifiable Robustness (CVPR 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.
