Research map / Adversarial robustness and security / Certified and empirical robustness
Overcoming the Convex Barrier for Simplex Inputs
Harkirat Singh, M. Pawan Kumar, Philip Torr, Krishnamurthy Dj Dvijotham
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 Simplex relaxation instead of Planet relaxation, both solved exactly with Gurobi (ablation) | Gurobi · MNIST OVAL Wide | verified accuracy: 31.7% -> 45.2% | Improved |
| Gurobi Simplex vs Gurobi Planet (ablation) | Gurobi · CIFAR-10 OVAL Wide | verified accuracy: 18.4% -> 29.4% | Improved |
| Simplex Verify instead of Opt-Lirpa Planet at same time | LiRPA solver · CIFAR-10 VNN Med | verified accuracy: 48.8% -> 59.4% | Improved |
| Simplex Verify instead of Opt-Lirpa Planet | LiRPA solver · CIFAR-10 VNN Big | verified accuracy: 60.0% -> 66.4% | Improved |
| Simplex Verify instead of Opt-Lirpa Planet | LiRPA solver · MNIST OVAL Wide | verified accuracy: 31.0% -> 44.6% | Improved |
| Simplex Verify instead of Opt-Lirpa Planet on SGD-trained model | LiRPA solver · CIFAR-10 OVAL Wide (SGD) | verified accuracy: 13.5% -> 22.4% | Improved |
| Simplex Verify instead of Opt-Lirpa Planet on small-margin Deep model | LiRPA solver · CIFAR-10 OVAL Deep | verified accuracy: 10.8% -> 13.4% | Improved |
| fast LiRPA solver for Simplex relaxation instead of Gurobi | Simplex Verify · MNIST OVAL Wide | verified accuracy: 45.2% -> 44.6% | No clear effect |
| fast LiRPA solver for Simplex relaxation instead of Gurobi | Simplex Verify · MNIST OVAL Wide | time per sample: 72.47s -> 0.04s | Improved |
| Simplex Verify vs Opt-Lirpa Planet runtime | LiRPA solvers · CIFAR-10 OVAL Deep | time per sample: 0.06s -> 0.05s | No clear effect |
| Simplex Verify instead of Opt-Lirpa Planet, text-perturbation, network 1 | ConcatBOW · Food-101 10-class | verified accuracy: 17.9% -> 42.2% | Improved |
| Simplex Verify instead of Opt-Lirpa Planet, text-perturbation, network 2 | ConcatBOW · Food-101 10-class | verified accuracy: 08.0% -> 32.8% | Improved |
Closest papers by meaning
- A Convex Relaxation Barrier to Tight Robustness Verification of Neural Networks (NeurIPS 2019 · similarity 0.92)
- The Convex Relaxation Barrier, Revisited: Tightened Single-Neuron Relaxations for Neural Network Verification (NeurIPS 2020 · similarity 0.91)
- Expressiveness of Multi-Neuron Convex Relaxations in Neural Network Certification (ICLR 2026 · similarity 0.90)
- Convex Formulations for Training Two-Layer ReLU Neural Networks (ICLR 2025 · similarity 0.87)
- Expressivity of ReLU-Networks under Convex Relaxations (ICLR 2024 · similarity 0.87)
- Probably Approximately Global Robustness Certification (ICML 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.
