Research map / Adversarial robustness and security / Certified and empirical robustness
Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing Their Input Gradients
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 |
|---|---|---|---|
| Regularize input gradients using certainty sensitivity | CNN · MNIST (JSMA) | human fooled: 2.0% -> 16.4% | Helped |
| Regularize input gradients using certainty sensitivity | CNN · MNIST (JSMA) | mistake reasonable: 26.0% -> 41.8% | Helped |
| Regularize input gradients using certainty sensitivity | CNN · SVHN (TGSM) | human fooled: 40.0% -> 46.3% | Helped |
| Regularize input gradients using certainty sensitivity | CNN · SVHN (TGSM) | mistake reasonable: 63.3% -> 81.5% | Helped |
| Regularize input gradients using certainty sensitivity instead of using no defense | CNN · MNIST (JSMA) | human fooled: 0.0% -> 16.4% | Helped |
| Regularize input gradients using certainty sensitivity instead of using no defense | CNN · SVHN (TGSM) | human fooled: 1.7% -> 46.3% | Helped |
| Increase gradient-regularization strength (ablation) | CNN · FGSM validation examples | accuracy | Helped |
| Over-regularize input gradients (ablation) | CNN · Normal and adversarial examples | accuracy | Hurt |
| Combine gradient regularization with adversarial training (ablation) | CNN · SVHN | robustness to white-box and normal black-box attacks | Helped |
Closest papers by meaning
- Towards Understanding the Regularization of Adversarial Robustness on Neural Networks (ICML 2020 · similarity 0.92)
- Proper Network Interpretability Helps Adversarial Robustness in Classification (ICML 2020 · similarity 0.91)
- Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples (ICML 2023 · similarity 0.91)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.89)
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.