Research map / Learning theory and algorithms / Sample complexity
Towards Evaluating and Training Verifiably Robust Neural Networks
Check 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 |
|---|---|---|---|
| Replace ReLU with ParamRamp (η decreases from 0.01 to 0) (ablation) | IBP-trained DM-large · CIFAR-10, ε=2/255 | IBP verified error (%): 55.19 -> 47.84 | Helped |
| Replace ReLU with ParamRamp (η decreases from 0.01 to 0) (ablation) | IBP-trained DM-large · CIFAR-10, ε=8/255 | IBP verified error (%): 70.81 -> 67.26 | Helped |
| Replace ReLU with ParamRamp (η decreases from 0.01 to 0) (ablation) | CROWN-IBP-trained DM-large · CIFAR-10, ε=8/255 | IBP verified error (%): 66.94 -> 65.19 | Helped |
| Replace ReLU with ParamRamp (η decreases from 0.01 to 0) (ablation) | IBP-trained DM-large · MNIST, ε=0.4 | IBP verified error (%): 14.80 -> 10.90 | Helped |
| Replace ReLU with ParamRamp (η decreases from 0.01 to 0) (ablation) | CROWN-IBP-trained DM-large · MNIST, ε=0.4 | CROWN-LBP verified error (%): 11.90 -> 10.61 | Helped |
| Use CROWN-LBP with tight rather than adaptive bounding lines (ablation) | IBP-trained CIFAR-10 classifier · CIFAR-10, ε=8/255 | verified error (%): 99.99 -> 69.98 | Helped |
| Use CROWN-IBP with tight rather than adaptive bounding lines (ablation) | IBP-trained CIFAR-10 classifier · CIFAR-10, ε=8/255 | verified error (%): 85.66 -> 70.01 | Helped |
| Use tight CROWN-LBP instead of IBP for verification | IBP-trained CIFAR-10 classifier · CIFAR-10, ε=8/255 | verified error (%): 70.10 -> 69.98 | Helped |
| Use adaptive CROWN-LBP instead of IBP for verification | IBP-trained CIFAR-10 classifier · CIFAR-10, ε=8/255 | verified error (%): 70.10 -> 99.99 | Hurt |
| Use ParamRamp (η decreases from 0.01 to 0) instead of ReLU (ablation) | IBP-trained DM-large · CIFAR-10, ε=2/255 | PGD attack error (%): 50.40 -> 42.74 | Helped |
Closest papers by meaning
- Towards Stable and Efficient Training of Verifiably Robust Neural Networks (ICLR 2020 · similarity 0.96)
- Understanding Certified Training with Interval Bound Propagation (ICLR 2024 · similarity 0.92)
- Beta-CROWN: Efficient Bound Propagation with Per-neuron Split Constraints for Neural Network Robustness Verification (NeurIPS 2021 · similarity 0.92)
- SDP-CROWN: Efficient Bound Propagation for Neural Network Verification with Tightness of Semidefinite Programming (ICML 2025 · similarity 0.92)
- On the Convergence of Certified Robust Training with Interval Bound Propagation (ICLR 2022 · similarity 0.90)
- Clip-and-Verify: Linear Constraint-Driven Domain Clipping for Accelerating Neural Network Verification (NeurIPS 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.