Research map / Adversarial robustness and security / Certified and empirical robustness
Automatic Perturbation Analysis for Scalable Certified Robustness and Beyond
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 |
|---|---|---|---|
| train with tight LiRPA (IBP+backward) bounds instead of IBP | certified training CNN-7+BN · CIFAR-10 eps=8/255 | verified error: 69.56% -> 66.62% | Helped |
| LiRPA bounds instead of IBP | certified training WideResNet · Tiny-ImageNet eps=1/255 | verified error: 85.15% -> 84.14% | Helped |
| LiRPA bounds instead of IBP | certified training · Downscaled ImageNet | verified error: 93.87% -> 91.27% | Helped |
| add loss fusion (ablation) | LiRPA training ResNeXt · CIFAR-10 | epoch time (s): 132.70 -> 55.84 | Helped |
| add loss fusion (ablation) | LiRPA training ResNeXt · CIFAR-10 | GPU memory (GB): 43.68 -> 18.58 | Helped |
| add loss fusion (ablation) | LiRPA training · Tiny-ImageNet | feasibility (OOM) | Helped |
| IBP+Backward instead of IBP for certified training | Transformer · SST-2 delta=6 | verified accuracy: 76.6 -> 77.4 | Helped |
| IBP+Backward instead of IBP | LSTM · SST-2 delta=6 | verified accuracy: 76.2 -> 76.6 | No clear effect |
| add backward pass to forward-mode verification | naturally trained Transformer · SST-2 delta=6 | verified accuracy: 35.0 -> 38.7 | Helped |
| train with LiRPA flatness objective | 3-layer MLP · MNIST/FashionMNIST subsets | test accuracy | Helped |
Closest papers by meaning
- Fast and Complete: Enabling Complete Neural Network Verification with Rapid and Massively Parallel Incomplete Verifiers (ICLR 2021 · similarity 0.87)
- Semidefinite relaxations for certifying robustness to adversarial examples (NeurIPS 2018 · similarity 0.86)
- GaLileo: General Linear Relaxation Framework for Tightening Robustness Certification of Transformers (AAAI 2024 · similarity 0.84)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.83)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.83)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · 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.