Research map / Architectures and efficiency / CNN design
Tightening Robustness Verification of Convolutional Neural Networks with Fine-Grained Linear Approximation
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 |
|---|---|---|---|
| Use fine-grained activation-function bounds instead of the CROWN approximation | DeepCert on MNIST 8-layer network with tanh · MNIST, 8-layer network, tanh, l1 | certified lower bound: 0.048 -> 0.185 | Helped |
| Use fine-grained activation-function bounds instead of the FROWN approximation | DeepCert on MNIST 8-layer network with tanh · MNIST, 8-layer network, tanh, l1 | certified lower bound: 0.053 -> 0.185 | Helped |
| Use fine-grained activation-function bounds instead of the CROWN approximation | DeepCert on MNIST 4-layer network with sigmoid · MNIST, 4-layer network, sigmoid, l1 | certified lower bound: 0.482 -> 0.821 | Helped |
| Use fine-grained activation-function bounds instead of the FROWN approximation | DeepCert on MNIST 4-layer network with sigmoid · MNIST, 4-layer network, sigmoid, l1 | certified lower bound: 0.634 -> 0.821 | Helped |
| Use fine-grained activation-function bounds rather than CNN-Cert's bounds | DeepCert on MNIST LeNet-5 with sigmoid · MNIST, LeNet-5, sigmoid, l∞ | certified lower bound: 0.013 -> 0.013 | No clear effect |
| Use fine-grained activation-function bounds rather than CNN-Cert's bounds | DeepCert on MNIST ResNet with 5 blocks and tanh · MNIST, ResNet 5 blocks, tanh, l1 | certified lower bound: 0.037 -> 0.065 | Helped |
| Use fine-grained activation-function bounds rather than FROWN's gradient-based search | DeepCert on MNIST 8-layer network with sigmoid · MNIST, 8-layer network, sigmoid, l∞ | certified lower bound: 0.111 -> 0.099 | Hurt |
| Use fine-grained activation-function bounds rather than FROWN's gradient-based search | DeepCert on MNIST 8-layer network with sigmoid · MNIST, 8-layer network, sigmoid, l2 | certified lower bound: 0.428 -> 0.407 | Hurt |
| Use fine-grained activation-function bounds instead of CROWN's midpoint tangent approximation | DeepCert on MNIST 8-layer network with tanh · MNIST, 8-layer network, tanh, l1 | certification time: 4.0 -> 16.9 | Hurt |
| Use fine-grained activation-function bounds instead of FROWN's gradient-based search | DeepCert on MNIST 8-layer network with tanh · MNIST, 8-layer network, tanh, l1 | certification time: 3737.7 -> 16.9 | Helped |
Closest papers by meaning
- Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation (CVPR 2024 · similarity 0.91)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.90)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.89)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.89)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.88)
- VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees (ICML 2024 · similarity 0.88)
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.