Research map / Learning theory and algorithms / Combinatorial and search optimization
Tight Neural Network Verification via Semidefinite Relaxations and Linear Reformulations
Jianglin Lan, Yang Zheng, Alessio Lomuscio
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 |
|---|---|---|---|
| Add RLT-generated linear cuts to LayerSDP | RLT-SDP · MNIST, MLP-Adv | certified robustness: 80 | 91 -> 88 | 94 | Improved |
| Add RLT-generated linear cuts to LayerSDP | RLT-SDP · MNIST, MLP-LP | certified robustness: 80 | 80 -> 80 | 80 | No clear effect |
| Add RLT-generated linear cuts to LayerSDP | RLT-SDP · MNIST, MLP-SDP | certified robustness: 80 | 84 -> 84 | 84 | Improved |
| Add RLT-generated linear cuts to LayerSDP | RLT-SDP · MNIST, 6 × 100 | certified robustness: 60 | 75 -> 82 | 90 | Improved |
| Add RLT-generated linear cuts to LayerSDP | RLT-SDP · MNIST, 9 × 100 | certified robustness: 22 | 35 -> 56 | 70 | Improved |
| Use RLT-SDP instead of IBP | RLT-SDP · MNIST, 8×1024-0.1 | certified robustness: 67 -> 82 | Improved |
| Use RLT-SDP instead of IBP | RLT-SDP · MNIST, 8×1024-0.3 | certified robustness: 7 -> 26 | Improved |
| Add a larger percentage of RLT linear cuts (ablation) | RLT-SDP · random ReLU networks, feasible output region | over-approximation tightness | Improved |
| Add more RLT linear cuts (ablation) | RLT-SDP · random ReLU networks, runtime | runtime increase | Hurt |
| Use the first 20% of RLT linear cuts (ablation) | RLT-SDP · random ReLU networks, feasible output region | over-approximation tightness | Improved |
Closest papers by meaning
- A Semidefinite Relaxation Based Branch-and-Bound Method for Tight Neural Network Verification (AAAI 2023 · similarity 0.94)
- SDP-CROWN: Efficient Bound Propagation for Neural Network Verification with Tightness of Semidefinite Programming (ICML 2025 · similarity 0.89)
- Expressivity of ReLU-Networks under Convex Relaxations (ICLR 2024 · similarity 0.88)
- Convex Formulations for Training Two-Layer ReLU Neural Networks (ICLR 2025 · similarity 0.87)
- Convex Relaxations of ReLU Neural Networks Approximate Global Optima in Polynomial Time (ICML 2024 · similarity 0.87)
- Clip-and-Verify: Linear Constraint-Driven Domain Clipping for Accelerating Neural Network Verification (NeurIPS 2025 · 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.
