Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Training with Bi-directional Likelihood Regularization for Visual Classification
Weitao Wan, Jiansheng Chen, Ming-Hsuan Yang
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 |
|---|---|---|---|
| add bi-directional likelihood regularization (ablation) | adversarial training · CIFAR-10 ResNet-32 | clean accuracy (%): 79.37 -> 86.32 | Improved |
| add bi-directional likelihood regularization (ablation) | adversarial training · CIFAR-10 ResNet-32 | PGD-10 accuracy (%): 44.92 -> 50.18 | Improved |
| add bi-directional likelihood regularization (ablation) | adversarial training · CIFAR-10 WideResNet-32 | clean accuracy (%): 87.30 -> 92.12 | Improved |
| add bi-directional likelihood regularization (ablation) | adversarial training · CIFAR-100 ResNet-32 | clean accuracy (%): 55.11 -> 67.34 | Improved |
| add bi-directional likelihood regularization (ablation) | adversarial training · MNIST PGD 100 steps 20 restarts | accuracy (%): 89.3 -> 92.7 | Improved |
| add bi-directional likelihood regularization (ablation) | adversarial training · MNIST clean | accuracy (%): 98.8 -> 99.3 | Improved |
| use likelihood weight lambda=0.001 versus 0 (plain adversarial training) (ablation) | adversarial training · CIFAR-10 ResNet-32 | clean accuracy (%): 79.37 -> 85.15 | Improved |
| increase lambda from 0.02 to 0.05 (ablation) | ATBLR · CIFAR-10 ResNet-32 | clean accuracy (%): 86.32 -> 86.09 | Hurt |
| train with stronger PGD-100 adversaries instead of PGD-10 (ablation) | ATBLR · CIFAR-10 ResNet-32 | clean accuracy gain over adv. training (points): 6.95 -> 9.04 | Improved |
| train with FGSM adversaries instead of PGD-10 (ablation) | ATBLR · CIFAR-10 ResNet-32 | PGD-10 accuracy (%): 50.18 -> 1.03 | Hurt |
| add bi-directional likelihood regularization to FGSM-trained model (ablation) | adversarial training · CIFAR-10 ResNet-32 trained with FGSM | clean accuracy (%): 87.40 -> 89.83 | Improved |
Closest papers by meaning
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Adversarial Distributional Training for Robust Deep Learning (NeurIPS 2020 · similarity 0.88)
- Robustness and Generalization via Generative Adversarial Training (ICCV 2021 · similarity 0.88)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.88)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · 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.
