Research map / Adversarial robustness and security / Certified and empirical robustness
A Unified Contrastive Energy-based Model for Understanding the Generative Ability of Adversarial Training
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 |
|---|---|---|---|
| Use MaxEnt sampling instead of PGD | Unsupervised CEM, ResNet18, unconditional sampling · CIFAR-10 | IS: 5.35 -> 8.24 | Helped |
| Use MaxEnt sampling instead of PGD | Unsupervised CEM, ResNet18, conditional sampling · CIFAR-10 | FID: 68.54 -> 36.44 | Helped |
| Use conditional rather than unconditional MaxEnt sampling | Unsupervised CEM, ResNet50 · CIFAR-10 | IS: 9.57 -> 9.61 | Helped |
| Add the positive logsumexp gradient to conditional sampling | Supervised CEM sampling · CIFAR-10 | IS: 9.77 -> 9.80 | Helped |
| Replace targeted attack sampling with conditional sampling | Supervised CEM sampling · CIFAR-10 | IS: 9.26 -> 9.77 | Helped |
| Add consistency regularization to AT | ResNet18 supervised adversarial training · CIFAR-10 | Adversarial accuracy: 52.2 -> 55.2 | Helped |
| Add consistency regularization to AT | WideResNet34 supervised adversarial training · CIFAR-10 | Adversarial accuracy: 53.6 -> 57.0 | Helped |
| Add consistency regularization to UAT | ResNet18 unsupervised adversarial training · CIFAR-10 | PGD20 adversarial accuracy: 21.4% to 24.6% | Helped |
| Add consistency regularization to UAT | ResNet18 unsupervised adversarial training · CIFAR-10 | Natural accuracy: 66.6% to 72.0% | Helped |
| Add consistency regularization to UAT | ResNet18 unsupervised adversarial training · CIFAR-10 | FGSM adversarial accuracy: 26.2% to 30.7% | Helped |
| Use unsupervised adversarial training instead of standard training | ResNet18 contrastive encoder · CIFAR-10 | PGD20 adversarial accuracy: 0.8% to 21.4% | Helped |
Closest papers by meaning
- Scalable Energy-Based Models via Adversarial Training: Unifying Discrimination and Generation (ICLR 2026 · similarity 0.91)
- Calibrating Energy-based Generative Adversarial Networks (ICLR 2017 · similarity 0.89)
- Robust Pre-Training by Adversarial Contrastive Learning (NeurIPS 2020 · similarity 0.88)
- Improving Adversarial Energy-Based Model via Diffusion Process (ICML 2024 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- DAT: Improving Adversarial Robustness via Generative Amplitude Mix-up in Frequency Domain (NeurIPS 2024 · 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.