Research map / Learning theory and algorithms / Prediction and decision losses
Efficient Statistical Assessment of Neural Network Corruption Robustness
Karim TIT, Teddy Furon, Mathias ROUSSET
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 Last Particle instead of DeepPoly | Last Particle · MNIST eps=0.015 | certified (%): 82 -> 99 | Improved |
| use Last Particle instead of DeepPoly | Last Particle · MNIST eps=0.03 | certified (%): 62 -> 97 | Improved |
| use Last Particle instead of DeepPoly | Last Particle · MNIST eps=0.06 | certified (%): 28 -> 93 | Improved |
| use Last Particle instead of DeepPoly | Last Particle · MNIST eps=0.1 | certified (%): 22 -> 85 | Improved |
| use Last Particle instead of DeepPoly | Last Particle · MNIST eps=0.015 | runtime (sec.): 5.69 -> 1.04 ± 0.005 | Improved |
| use Last Particle instead of DeepPoly | Last Particle · MNIST eps=0.1 | runtime (sec.): 8.84 -> 0.96 ± 0.02 | Improved |
| increase epsilon from 0.015 to 0.1 | Last Particle · MNIST | runtime (sec.): 1.04 ± 0.005 -> 0.96 ± 0.02 | Improved |
| increase epsilon from 0.015 to 0.1 | ERAN DeepPoly · MNIST | runtime (sec.): 5.69 -> 8.84 | Hurt |
| increase number of particles N from 2 to 20 | Last Particle · idealized linear case | score function calls: 167 -> 1470 | Hurt |
| increase epsilon from 0.02 to 0.06 | Last Particle · ImageNet ResNet50 | certified (%): 81 -> 59 | Hurt |
| increase epsilon from 0.02 to 0.06 | Last Particle · ImageNet MobileNet | avg runtime (sec.): 20.78 ± 0.74 -> 14.5 ± 0.11 | Improved |
| use Last Particle vs ERAN on ACAS Xu properties | Last Particle · ACAS Xu | number of cases certified: 107 certified, 9 false positives | Mixed |
Closest papers by meaning
- A Statistical Approach to Assessing Neural Network Robustness (ICLR 2019 · similarity 0.92)
- PROVEN: Verifying Robustness of Neural Networks with a Probabilistic Approach (ICML 2019 · similarity 0.89)
- How Sampling Impacts the Robustness of Stochastic Neural Networks (NeurIPS 2022 · similarity 0.87)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.86)
- Halt Fast! Early Stopping for Certified Robustness (NeurIPS 2026 · similarity 0.85)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.85)
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.
