Research map / Adversarial robustness and security / Certified and empirical robustness
RecurJac: An Efficient Recursive Algorithm for Bounding Jacobian Matrix of Neural Networks and Its Applications
Huan Zhang, Pengchuan Zhang, Cho-Jui Hsieh
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 |
|---|---|---|---|
| Use RecurJac instead of FastLip | MNIST 3-layer network · runner-up target, undefended | certified lower bound for ℓ∞ distortion: 0.01802 -> 0.02256 | Improved |
| Use RecurJac instead of FastLip | MNIST 3-layer network · runner-up target, adversarial training | certified lower bound for ℓ∞ distortion: 0.09639 -> 0.11573 | Improved |
| Use RecurJac instead of FastLip | MNIST 3-layer network · random target, undefended | certified lower bound for ℓ∞ distortion: 0.02374 -> 0.02870 | Improved |
| Use RecurJac instead of FastLip | MNIST 3-layer network · random target, adversarial training | certified lower bound for ℓ∞ distortion: 0.11753 -> 0.13753 | Improved |
| Use RecurJac instead of FastLip | MNIST 3-layer network · least-likely target, undefended | certified lower bound for ℓ∞ distortion: 0.02720 -> 0.03205 | Improved |
| Use RecurJac instead of FastLip | MNIST 3-layer network · least-likely target, adversarial training | certified lower bound for ℓ∞ distortion: 0.14067 -> 0.16153 | Improved |
| Use RecurJac instead of FastLip | MNIST 4-layer network · runner-up target, undefended | certified lower bound for ℓ∞ distortion: 0.01602 -> 0.02104 | Improved |
| Use RecurJac instead of FastLip | MNIST 4-layer network · runner-up target, adversarial training | certified lower bound for ℓ∞ distortion: 0.04232 -> 0.07350 | Improved |
| Use RecurJac instead of FastLip | MNIST 4-layer network · random target, undefended | certified lower bound for ℓ∞ distortion: 0.01882 -> 0.02399 | Improved |
| Use RecurJac instead of FastLip | MNIST 4-layer network · random target, adversarial training | certified lower bound for ℓ∞ distortion: 0.05267 -> 0.08603 | Improved |
| Use RecurJac instead of FastLip | MNIST 4-layer network · least-likely target, undefended | certified lower bound for ℓ∞ distortion: 0.02018 -> 0.02519 | Improved |
| Use RecurJac instead of FastLip | MNIST 4-layer network · least-likely target, adversarial training | certified lower bound for ℓ∞ distortion: 0.06417 -> 0.09863 | Improved |
Closest papers by meaning
- Efficiently Computing Local Lipschitz Constants of Neural Networks via Bound Propagation (NeurIPS 2022 · similarity 0.89)
- Exactly Computing the Local Lipschitz Constant of ReLU Networks (NeurIPS 2020 · similarity 0.89)
- Training Certifiably Robust Neural Networks with Efficient Local Lipschitz Bounds (NeurIPS 2021 · similarity 0.88)
- ECLipsE: Efficient Compositional Lipschitz Constant Estimation for Deep Neural Networks (NeurIPS 2024 · similarity 0.88)
- Efficient and Accurate Estimation of Lipschitz Constants for Deep Neural Networks (NeurIPS 2019 · similarity 0.88)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.86)
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.
