Research map / Learning theory and algorithms / Combinatorial and search optimization
Enabling certification of verification-agnostic networks via memory-efficient semidefinite programming
Sumanth Dathathri, Krishnamurthy Dvijotham, Alexey Kurakin, Aditi Raghunathan, Jonathan Uesato, Rudy R Bunel, Shreya Shankar, Jacob Steinhardt, Ian Goodfellow, Percy Liang, Pushmeet Kohli
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 SDP-FO instead of LP relaxation | verification-agnostic CNN-A-Adv · MNIST eps=0.1 | verified accuracy (%): 0.4 -> 87.8 | Improved |
| Use SDP-FO instead of LP relaxation | CNN-A-Mix · CIFAR-10 eps=2/255 | verified accuracy (%): 5.8 -> 39.6 | Improved |
| Use SDP-FO instead of LP relaxation | MLP-LP (verification-aware) · MNIST eps=0.1 | verified accuracy (%): 79.4 -> 80.2 | Improved |
| Use SDP-FO instead of MIP | MLP-Adv-B · MNIST eps=0.1 | verified accuracy (%): 34.4 -> 79.2 | Improved |
| Use SDP-FO instead of SDP-IP interior point | MLP-Adv · MNIST eps=0.1 | verified accuracy (%): 82 -> 91.0 | Improved |
| Scale network from CNN-A to larger CNN-B | SDP-FO · CIFAR-10 eps=2/255 | verified accuracy (%): 39.6 -> 32.8 | Hurt |
| Use CROWN precomputed bounds instead of interval arithmetic bounds (ablation) | SDP-FO · MNIST MLPs | bound gap | Improved |
| Use SDP-FO instead of IBP | VAE decoder verification · MNIST latent perturbation | perturbation radius verified for 50%: 0.01 -> 0.5 | Improved |
| Use Lanczos instead of exact eigendecomposition during optimization (ablation) | SDP-FO · large CNN speed | time per iteration: 5 minutes -> 0.2s per iteration | Improved |
Closest papers by meaning
- Iteratively Enhanced Semidefinite Relaxations for Efficient Neural Network Verification (AAAI 2023 · similarity 0.92)
- SDP-CROWN: Efficient Bound Propagation for Neural Network Verification with Tightness of Semidefinite Programming (ICML 2025 · similarity 0.89)
- Verifying Properties of Binary Neural Networks Using Sparse Polynomial Optimization (ICLR 2025 · similarity 0.88)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.87)
- Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation (CVPR 2024 · similarity 0.87)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.87)
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.
