Research map / Adversarial robustness and security / Certified and empirical robustness
Absum: Simple Regularization Method for Reducing Structural Sensitivity of Convolutional Neural Networks
Sekitoshi Kanai, Yasutoshi Ida, Yasuhiro Fujiwara, Masanori Yamada, Shuichi Adachi
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 Absum instead of standard weight decay | CNNs · MNIST, FMNIST, and SVHN under SFA | average robust accuracy: 98.59 -> 98.64; 83.09 -> 83.11; 93.20 -> 93.34 | Improved |
| Use Absum instead of L1 regularization | CNNs · MNIST, FMNIST, CIFAR10, CIFAR100, and SVHN under SFA | average robust accuracy: 98.48 -> 98.64; 82.49 -> 83.11; 66.44 -> 79.05; 38.99 -> 48.69; 91.14 -> 93.34 | Mixed |
| Use spectral norm constraint instead of Absum | CNNs · CIFAR10 and CIFAR100 under SFA | average robust accuracy: 79.05 -> 85.57; 48.69 -> 60.42 | Improved |
| Use Absum instead of spectral norm constraint | CNNs · MNIST, FMNIST, and SVHN under SFA | average robust accuracy: 98.55 -> 98.64; 82.60 -> 83.11; 93.20 -> 93.34 | Improved |
| Use Absum instead of weight decay | CNNs · MNIST, FMNIST, and SVHN under transferred PGD | robust accuracy: 48.94 -> 76.34; 3.46 -> 30.08; 40.49 -> 49.11 | Improved |
| Use Absum instead of L1 regularization | CNNs · SVHN under transferred PGD (ε=4/255) | robust accuracy: 52.79 -> 49.11 | Hurt |
| Use Absum instead of weight decay | CNNs · CIFAR10 and CIFAR100 under transferred PGD (ε=4/255) | robust accuracy: 18.48 -> 26.29; 17.40 -> 18.57 | Improved |
| Use Absum instead of L1 regularization | CNNs · SVHN under High-Frequency attack | robust accuracy: 17.03 -> 52.95 | Improved |
| Use Absum instead of weight decay | CNNs · CIFAR10 and CIFAR100 under High-Frequency attack | robust accuracy: 52.82 -> 64.51; 36.15 -> 41.44 | Improved |
| Add Absum regularization to adversarial training | CNNs trained with PGD adversarial training · MNIST and FMNIST | robust accuracy | Improved |
| Add Absum regularization to adversarial training | CNNs trained with PGD adversarial training · CIFAR10 and CIFAR100 | robust accuracy | Mixed |
| Add Absum regularization to naive training | CNNs without adversarial training · PGD attacks | robust accuracy | No clear effect |
Closest papers by meaning
- Large Norms of CNN Layers Do Not Hurt Adversarial Robustness (AAAI 2021 · similarity 0.89)
- Characterizing Model Robustness via Natural Input Gradients (ECCV 2024 · similarity 0.88)
- Improving DNN Robustness to Adversarial Attacks using Jacobian Regularization (ECCV 2018 · similarity 0.88)
- Adversarial Robustness through Local Linearization (NeurIPS 2019 · similarity 0.88)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.86)
- First Line of Defense: A Robust First Layer Mitigates Adversarial Attacks (AAAI 2025 · 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.
