Adaptive Diffusion Guidance via Stochastic Optimal Control
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914435629580288 |
|---|---|
| author | Azangulov, Iskander Potaptchik, Peter Li, Qinyu Aamari, Eddie Deligiannidis, George Rousseau, Judith |
| author_facet | Azangulov, Iskander Potaptchik, Peter Li, Qinyu Aamari, Eddie Deligiannidis, George Rousseau, Judith |
| contents | Guidance is a cornerstone of modern diffusion models, playing a pivotal role in conditional generation and enhancing the quality of unconditional samples. However, current approaches to guidance scheduling--determining the appropriate guidance weight--are largely heuristic and lack a solid theoretical foundation. This work addresses these limitations on two fronts. First, we provide a theoretical formalization that precisely characterizes the relationship between guidance strength and classifier confidence. Second, building on this insight, we introduce a stochastic optimal control framework that casts guidance scheduling as an adaptive optimization problem. In this formulation, guidance strength is not fixed but dynamically selected based on time, the current sample, and the conditioning class, either independently or in combination. By solving the resulting control problem, we establish a principled foundation for more effective guidance in diffusion models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_19367 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Adaptive Diffusion Guidance via Stochastic Optimal Control Azangulov, Iskander Potaptchik, Peter Li, Qinyu Aamari, Eddie Deligiannidis, George Rousseau, Judith Machine Learning Guidance is a cornerstone of modern diffusion models, playing a pivotal role in conditional generation and enhancing the quality of unconditional samples. However, current approaches to guidance scheduling--determining the appropriate guidance weight--are largely heuristic and lack a solid theoretical foundation. This work addresses these limitations on two fronts. First, we provide a theoretical formalization that precisely characterizes the relationship between guidance strength and classifier confidence. Second, building on this insight, we introduce a stochastic optimal control framework that casts guidance scheduling as an adaptive optimization problem. In this formulation, guidance strength is not fixed but dynamically selected based on time, the current sample, and the conditioning class, either independently or in combination. By solving the resulting control problem, we establish a principled foundation for more effective guidance in diffusion models. |
| title | Adaptive Diffusion Guidance via Stochastic Optimal Control |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2505.19367 |