Research map / Adversarial robustness and security / Certified and empirical robustness
Exploring the Vulnerability of Deep Neural Networks: A Study of Parameter Corruption
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 |
|---|---|---|---|
| Add virtual gradient-based corruption training | ResNet-34 · Tiny-ImageNet | accuracy: 66.56 -> 67.06 (+0.50) | Helped |
| Add virtual gradient-based corruption training | ResNet-34 · CIFAR-10 | accuracy: 94.26 -> 96.12 (+1.86) | Helped |
| Add virtual gradient-based corruption training | LSTM · PTB-LM | PPL: 70.09 -> 68.28 (-1.81) | Helped |
| Add gradient regularization | MLP · PTB-Parsing | UAS: 87.26 -> 87.89 (+0.63) | Helped |
| Add virtual gradient-based corruption training | Transformer · De-En | BLEU: 35.33 -> 35.69 (+0.36) | Helped |
| Train with virtual corruption | ResNet-34 · Tiny-ImageNet, corruption ε=0.2 | accuracy: 41.79 -> 52.51 | Helped |
| Train with virtual corruption | ResNet-34 · CIFAR-10, corruption ε=0.2 | accuracy: 71.44 -> 87.92 | Helped |
| Train with virtual corruption | LSTM · PTB-LM, corruption ε=14 | PPL: 240.99 -> 69.61 | Helped |
| Train with virtual corruption | MLP · PTB-Parsing, corruption ε=0.05 | UAS: 72.32 -> 82.06 | Helped |
| Train with virtual corruption | Transformer · De-En, corruption ε=1 | BLEU: 11.17 -> 31.25 | Helped |
| Use gradient-based instead of random corruption | ResNet-34 · ImageNet, ε=10⁻² | accuracy: 30.3 -> 0.1 | Hurt |
| Use gradient-based instead of random corruption | ResNet-34 · CIFAR-10, ε=10⁻² | accuracy: 75.1 -> 10.0 | Hurt |
Closest papers by meaning
- Feature Collapse Under Corruption: An Entropy Perspective on Robust Neural Networks (ICML 2026 · similarity 0.89)
- First-Order Adversarial Vulnerability of Neural Networks and Input Dimension (ICML 2019 · similarity 0.89)
- Measuring Neural Net Robustness with Constraints (NeurIPS 2016 · similarity 0.88)
- Nasty Adversarial Training: A Probability Sparsity Perspective for Robustness Enhancement (ICLR 2026 · similarity 0.88)
- Neural Architecture Design and Robustness: A Dataset (ICLR 2023 · similarity 0.88)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · 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.