Research map / Adversarial robustness and security / Certified and empirical robustness
Concise Explanations of Neural Networks using Adversarial Training
Prasad Chalasani, Jiefeng Chen, Amrita Roy Chowdhury, Xi Wu, Somesh Jha
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 ℓ∞-bounded adversarial training instead of natural training with ℓ1 regularization | CNNs · MNIST | IG attribution sparseness improvement (dG): 0.004 vs 0.06 | Improved |
| Use ℓ∞-bounded adversarial training instead of natural training with ℓ1 regularization | CNNs · Fashion-MNIST | attribution sparseness improvement (dG): 0.008 vs 0.06; 0.003 vs 0.08 | Improved |
| Use ℓ∞-bounded adversarial training instead of natural training with ℓ1 regularization | CNN · CIFAR-10 | IG attribution sparseness improvement (dG): 0.022 vs 0.081 | Improved |
| Use ℓ∞-bounded adversarial training instead of natural training with ℓ1 regularization | logistic regression · Mushroom | IG attribution sparseness improvement (dG): 0.06 vs 0.06 | No clear effect |
| Use ℓ∞-bounded adversarial training instead of natural training with ℓ1 regularization | logistic regression · Spambase | IG attribution sparseness improvement (dG): 0.15 vs 0.17 | Improved |
| Use ℓ∞-bounded adversarial training | 1-layer models with the specified loss and data assumptions · theoretical analysis | IG attribution sparseness | Improved |
| Add an IG-stability penalty to natural training | 1-layer models with the specified loss · theoretical analysis | adversarial objective equivalence | No clear effect |
Closest papers by meaning
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.89)
- Attribution Explanations for Deep Neural Networks: A Theoretical Perspective (TPAMI 2026 · similarity 0.89)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 2025 · similarity 0.89)
- Interpreting Deep Neural Networks with Relative Sectional Propagation by Analyzing Comparative Gradients and Hostile Activations (AAAI 2021 · similarity 0.87)
- Interpreting Adversarially Trained Convolutional Neural Networks (ICML 2019 · similarity 0.87)
- Identifying and Understanding Cross-Class Features in Adversarial Training (ICML 2025 · similarity 0.87)
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.
