Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Verifying Robustness of Neural Networks Against A Family of Semantic Perturbations
Check 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 |
|---|---|---|---|
| Add semantic perturbation layers instead of directly converting semantic changes to pixel-wise weighted ε bounds | Semantify-NN · CIFAR-10, Hue; MLP 6 × 2048 | certified bound: 0.028 -> 0.347 | Helped |
| Add explicit input splitting to SPL (ablation) | SPL · CIFAR-10, Hue; MLP 6 × 2048 | certified bound: 0.347 -> 0.974 | Helped |
| Add semantic perturbation layers instead of directly converting semantic changes to pixel-wise weighted ε bounds | Semantify-NN · CIFAR-10, Saturation; MLP 6 × 2048 | certified bound: 0.004 -> 0.101 | Helped |
| Add explicit input splitting to SPL (ablation) | SPL · CIFAR-10, Saturation; MLP 6 × 2048 | certified bound: 0.101 -> 0.314 | Helped |
| Add semantic perturbation layers instead of directly converting semantic changes to pixel-wise weighted ε bounds | Semantify-NN · CIFAR-10, Lightness; MLP 6 × 2048 | certified bound: 0.001 -> 0.047 | Helped |
| Add explicit input splitting to SPL (ablation) | SPL · CIFAR-10, Lightness; MLP 6 × 2048 | certified bound: 0.047 -> 0.244 | Helped |
| Increase implicit splits from 1 to 10 (ablation) | SPL rotation verifier · MNIST, Rotation; MLP 2 × 1024 | certified bound (degrees): 0.627 -> 2.515 | Helped |
| Use 100 implicit splits plus explicit intervals of 0.5° (ablation) | SPL rotation verifier · MNIST, Rotation; MLP 2 × 1024 | certified bound (degrees): 2.515 -> 46.24 | Helped |
| Increase implicit splits from 1 to 10 (ablation) | SPL rotation verifier · CIFAR-10, Rotation; MLP 5 × 2048 | certified bound (degrees): 0.006 -> 0.033 | Helped |
| Use 100 implicit splits plus explicit intervals of 0.5° (ablation) | SPL rotation verifier · CIFAR-10, Rotation; MLP 5 × 2048 | certified bound (degrees): 0.033 -> 14.81 | Helped |
Closest papers by meaning
- Towards the first adversarially robust neural network model on MNIST (ICLR 2019 · similarity 0.88)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.86)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.86)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.85)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.85)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.85)
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.