Research map / Adversarial robustness and security / Certified and empirical robustness
Understanding the Limitations of Conditional Generative Models
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 a conditional split prior instead of the standard class-conditional prior (ablation) | GLOW · MNIST | accuracy (%): 99.0 -> 99.3 | Helped |
| Use a conditional split prior instead of the standard class-conditional prior (ablation) | GLOW · MNIST | bits/dim: 1.10 -> 1.00 | Helped |
| Use a conditional split prior instead of the standard class-conditional prior (ablation) | GLOW · CIFAR10 | accuracy (%): 83.2 -> 84.0 | Helped |
| Use a conditional split prior instead of the standard class-conditional prior (ablation) | GLOW · CIFAR10 | bits/dim: 3.54 -> 3.53 | Helped |
| Add a classification loss term (ablation) | GLOW · MNIST | accuracy (%): 96.9 -> 99.0 | Helped |
| Add a classification loss term (ablation) | GLOW · MNIST | bits/dim: 0.95 -> 1.10 | Hurt |
| Add a classification loss term (ablation) | GLOW · CIFAR10 | accuracy (%): 56.8 -> 83.2 | Helped |
| Add a classification loss term (ablation) | GLOW · CIFAR10 | bits/dim: 3.47 -> 3.54 | Hurt |
| Use the modified boundary attack that also targets the detector | Reweighted and Split GLOW · CIFAR10 | successful attacks (%): 100% successful for both models | Hurt |
| Use the modified boundary attack that also targets the detector | Reweighted and Split GLOW · MNIST | successful attacks (%): 0% successful for both models | Helped |
| Interpolate between images from different classes | GLOW · CIFAR10 | interpolations above detection threshold (%): 100% | Hurt |
| Interpolate between images from different classes | GLOW · MNIST | interpolations above detection threshold (%): 0.5% | Helped |
Closest papers by meaning
- Robust Learning Meets Generative Models: Can Proxy Distributions Improve Adversarial Robustness? (ICLR 2022 · similarity 0.89)
- Adversarial vulnerability for any classifier (NeurIPS 2018 · similarity 0.89)
- Classification Accuracy Score for Conditional Generative Models (NeurIPS 2019 · similarity 0.88)
- Improving Robustness using Generated Data (NeurIPS 2021 · similarity 0.88)
- Generative Classifiers Avoid Shortcut Solutions (ICLR 2025 · similarity 0.87)
- A Likelihood Based Approach to Distribution Regression Using Conditional Deep Generative Models (ICML 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.