Research map / Adversarial robustness and security / Certified and empirical robustness
PROVEN: Verifying Robustness of Neural Networks with a Probabilistic Approach
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 probabilistic certification instead of worst-case certification | ReLU network, MNIST 2×[20] · MNIST, bounded independent inputs | certified ε: 0.02722 -> 0.04394 | Helped |
| Add adaptive activation bounds (ablation) | PROVEN on ReLU network, MNIST 2×[20] · MNIST, bounded independent inputs | certified ε: 0.04394 -> 0.04912 | Helped |
| Use probabilistic certification with adaptive activation bounds instead of worst-case certification | ReLU network, MNIST 2×[1024] · MNIST, bounded independent inputs | certified ε: 0.03158 -> 0.05560 | Helped |
| Use probabilistic certification with adaptive activation bounds instead of worst-case certification | ReLU network, CIFAR 5×[2048] · CIFAR, bounded independent inputs | certified ε: 0.00228 -> 0.00264 | Helped |
| Use probabilistic certification instead of worst-case certification | tanh network, MNIST 2×[1024] · MNIST, bounded independent inputs | certified ε: 0.02232 -> 0.02915 | Helped |
| Use probabilistic certification instead of worst-case certification | sigmoid network, MNIST 2×[1024] · MNIST, bounded independent inputs | certified ε: 0.02785 -> 0.03285 | Helped |
| Use probabilistic certification instead of worst-case certification | arctan network, MNIST 2×[1024] · MNIST, bounded independent inputs | certified ε: 0.02105 -> 0.02796 | Helped |
| Increase the number of sampled images from 10 to 100 when estimating mean certification (ablation) | PROVEN and worst-case certification, MNIST 3×[1024] ReLU with adaptive bounds · MNIST, 100 random trials | standard deviation of improvement: 1.12% -> 0.43% | Helped |
Closest papers by meaning
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.91)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.90)
- VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees (ICML 2024 · similarity 0.89)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.89)
- Halt Fast! Early Stopping for Certified Robustness (NeurIPS 2026 · similarity 0.88)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 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.