Research map / Learning theory and algorithms / Combinatorial and search optimization
Towards Fast Computation of Certified Robustness for ReLU Networks
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 Fast-Lin rather than the global operator-norm bound | ReLU robustness certification · MNIST 2×[1024], ℓ∞, random target | certified lower bound: 0.00263 -> 0.03083 | Helped |
| Use Fast-Lin rather than the global operator-norm bound | ReLU robustness certification · CIFAR 7×[1024], ℓ2, random target | certified lower bound: 0.00029 -> 0.05938 | Helped |
| Use Fast-Lin instead of LP | MNIST 2×[1024] robustness certification · MNIST 2×[1024], ℓ∞, random target | running time per image: 20.8 s -> 156 ms | Helped |
| Apply adversarial training with ε = 0.3 | MNIST 3×[1024] with Fast-Lin · MNIST 3×[1024], runner-up target, ℓ∞ | certified lower bound: 0.01826 -> 0.14730 | Helped |
| Apply defensive distillation with temperature = 100 | MNIST 3×[1024] with Fast-Lin · MNIST 3×[1024], runner-up target, ℓ∞ | certified lower bound: 0.01826 -> 0.02724 | Helped |
Closest papers by meaning
- Training for Faster Adversarial Robustness Verification via Inducing ReLU Stability (ICLR 2019 · similarity 0.92)
- Efficient Neural Network Robustness Certification with General Activation Functions (NeurIPS 2018 · similarity 0.90)
- Evaluating Robustness of Neural Networks with Mixed Integer Programming (ICLR 2019 · similarity 0.90)
- The Cost of Robustness: Tighter Bounds on Parameter Complexity for Robust Memorization in ReLU Nets (NeurIPS 2025 · similarity 0.89)
- Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation (CVPR 2024 · similarity 0.88)
- A Rescaling-Invariant Lipschitz Bound Based on Path-Metrics for Modern ReLU Network Parameterizations (ICML 2025 · 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.