Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866915762067734528 |
|---|---|
| author | Jia, Jing Yuan, Wei Liu, Sifan Shen, Liyue Wang, Guanyang |
| author_facet | Jia, Jing Yuan, Wei Liu, Sifan Shen, Liyue Wang, Guanyang |
| contents | Can a diffusion model trained on bedrooms recover human faces? Diffusion models are widely used as priors for inverse problems, but standard approaches usually assume a high-fidelity model trained on data that closely match the unknown signal. In practice, one often must use a mismatched or low-fidelity diffusion prior. Surprisingly, these weak priors often perform nearly as well as full-strength, in-domain baselines. We study when and why inverse solvers are robust to weak diffusion priors. Through extensive experiments, we find that weak priors succeed when measurements are highly informative (e.g., many observed pixels), and we identify regimes where they fail. Our theory, based on Bayesian consistency, gives conditions under which high-dimensional measurements make the posterior concentrate near the true signal. These results provide a principled justification on when weak diffusion priors can be used reliably. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_22443 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance Jia, Jing Yuan, Wei Liu, Sifan Shen, Liyue Wang, Guanyang Machine Learning Computer Vision and Pattern Recognition Computation Can a diffusion model trained on bedrooms recover human faces? Diffusion models are widely used as priors for inverse problems, but standard approaches usually assume a high-fidelity model trained on data that closely match the unknown signal. In practice, one often must use a mismatched or low-fidelity diffusion prior. Surprisingly, these weak priors often perform nearly as well as full-strength, in-domain baselines. We study when and why inverse solvers are robust to weak diffusion priors. Through extensive experiments, we find that weak priors succeed when measurements are highly informative (e.g., many observed pixels), and we identify regimes where they fail. Our theory, based on Bayesian consistency, gives conditions under which high-dimensional measurements make the posterior concentrate near the true signal. These results provide a principled justification on when weak diffusion priors can be used reliably. |
| title | Weak Diffusion Priors Can Still Achieve Strong Inverse-Problem Performance |
| topic | Machine Learning Computer Vision and Pattern Recognition Computation |
| url | https://arxiv.org/abs/2601.22443 |