GradCheck: Analyzing classifier guidance gradients for conditional diffusion sampling
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911931958296576 |
|---|---|
| author | Vaeth, Philipp Fruehwald, Alexander M. Paassen, Benjamin Gregorova, Magda |
| author_facet | Vaeth, Philipp Fruehwald, Alexander M. Paassen, Benjamin Gregorova, Magda |
| contents | To sample from an unconditionally trained Denoising Diffusion Probabilistic Model (DDPM), classifier guidance adds conditional information during sampling, but the gradients from classifiers, especially those not trained on noisy images, are often unstable. This study conducts a gradient analysis comparing robust and non-robust classifiers, as well as multiple gradient stabilization techniques. Experimental results demonstrate that these techniques significantly improve the quality of class-conditional samples for non-robust classifiers by providing more stable and informative classifier guidance gradients. The findings highlight the importance of gradient stability in enhancing the performance of classifier guidance, especially on non-robust classifiers. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_17399 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | GradCheck: Analyzing classifier guidance gradients for conditional diffusion sampling Vaeth, Philipp Fruehwald, Alexander M. Paassen, Benjamin Gregorova, Magda Machine Learning To sample from an unconditionally trained Denoising Diffusion Probabilistic Model (DDPM), classifier guidance adds conditional information during sampling, but the gradients from classifiers, especially those not trained on noisy images, are often unstable. This study conducts a gradient analysis comparing robust and non-robust classifiers, as well as multiple gradient stabilization techniques. Experimental results demonstrate that these techniques significantly improve the quality of class-conditional samples for non-robust classifiers by providing more stable and informative classifier guidance gradients. The findings highlight the importance of gradient stability in enhancing the performance of classifier guidance, especially on non-robust classifiers. |
| title | GradCheck: Analyzing classifier guidance gradients for conditional diffusion sampling |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2406.17399 |