Research map / Adversarial robustness and security / Black-box and transfer attacks
Improving Adversarial Robustness via Probabilistically Compact Loss with Logit Constraints
Xin Li, Xiangrui Li, Deng Pan, Dongxiao Zhu
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 |
|---|---|---|---|
| Replace CE with PC loss and logit constraints | VGG-13 · CIFAR-100 PGD, 0.005 | accuracy: 24.4 -> 56.7 | Improved |
| Replace CE with PC loss and logit constraints | VGG-13 · Tiny ImageNet PGD, 0.005 | accuracy: 11.3 -> 24.8 | Improved |
| Replace CE with PC loss and logit constraints | LeNet-5 · MNIST PGD, 0.1 | accuracy: 49.0 -> 72.3 | Improved |
| Replace CE with PC loss and logit constraints | VGG-13 · CIFAR-10 PGD, 0.04 | accuracy: 0.0 -> 27.6 | Improved |
| Replace CE with PC loss and logit constraints | LeNet-5 · MNIST clean | accuracy: 99.2 -> 99.2 | No clear effect |
| Replace CE with PC loss and logit constraints | VGG-13 · CIFAR-10 clean | accuracy: 91.6 -> 91.2 | No clear effect |
| Replace GCE with PC loss and logit constraints | ResNet-56 · CIFAR-10 FGSM, 0.04 | accuracy: 41.2 -> 58.4 | Improved |
| Replace GCE with PC loss and logit constraints | LeNet-5 · MNIST FGSM, 0.1 | accuracy: 87.7 -> 80.5 | Hurt |
Closest papers by meaning
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness (ICLR 2020 · similarity 0.89)
- Expressive Losses for Verified Robustness via Convex Combinations (ICLR 2024 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.88)
- Enhancing Noise-Robust Losses for Large-Scale Noisy Data Learning (AAAI 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.
