Research map / Adversarial robustness and security / Certified and empirical robustness
Maximum-Entropy Adversarial Data Augmentation for Improved Generalization and Robustness
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 maximum-entropy regularization to ADA (ablation) | ADA with LeNet trained on MNIST · SVHN | accuracy (%): 35.70 ± 2.00 -> 42.00 ± 1.74 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with LeNet trained on MNIST · MNIST-M | accuracy (%): 58.65 ± 1.72 -> 63.98 ± 1.82 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with LeNet trained on MNIST · SYN | accuracy (%): 47.18 ± 0.61 -> 49.80 ± 1.74 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with LeNet trained on MNIST · USPS | accuracy (%): 80.40 ± 1.70 -> 79.10 ± 1.03 | Hurt |
| use a Bayesian neural network to estimate predictive uncertainty (ablation) | ME-ADA with LeNet trained on MNIST · Average across SVHN, MNIST-M, SYN, USPS | accuracy (%): 58.72 ± 1.12 -> 59.32 ± 0.82 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with AlexNet · PACS Art | accuracy (%): 64.3 -> 67.1 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with AlexNet · PACS Cartoon | accuracy (%): 69.8 -> 69.9 | No clear effect |
| add maximum-entropy regularization to ADA (ablation) | ADA with AlexNet · PACS Photo | accuracy (%): 85.1 -> 88.6 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with AlexNet · PACS Sketch | accuracy (%): 60.4 -> 63.0 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with AlexNet · PACS average | accuracy (%): 69.9 -> 72.2 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with AllConvNet, DenseNet, WideResNet, and ResNeXt · CIFAR-10-C average | accuracy (%): 75.1 -> 80.5 | Helped |
| add maximum-entropy regularization to ADA (ablation) | ADA with AllConvNet, DenseNet, WideResNet, and ResNeXt · CIFAR-100-C average | accuracy (%): 48.6 -> 52.3 | Helped |
Closest papers by meaning
- Revisiting Hilbert-Schmidt Information Bottleneck for Adversarial Robustness (NeurIPS 2021 · similarity 0.90)
- Boosting Adversarial Training via Fisher-Rao Norm-based Regularization (CVPR 2024 · similarity 0.88)
- ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck (ICML 2025 · similarity 0.88)
- Annealing Self-Distillation Rectification Improves Adversarial Training (ICLR 2024 · similarity 0.88)
- Revisiting Adversarial Training Under Long-Tailed Distributions (CVPR 2024 · similarity 0.87)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 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.