D-Flow: Differentiating through Flows for Controlled Generation
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909262650802176 |
|---|---|
| author | Ben-Hamu, Heli Puny, Omri Gat, Itai Karrer, Brian Singer, Uriel Lipman, Yaron |
| author_facet | Ben-Hamu, Heli Puny, Omri Gat, Itai Karrer, Brian Singer, Uriel Lipman, Yaron |
| contents | Taming the generation outcome of state of the art Diffusion and Flow-Matching (FM) models without having to re-train a task-specific model unlocks a powerful tool for solving inverse problems, conditional generation, and controlled generation in general. In this work we introduce D-Flow, a simple framework for controlling the generation process by differentiating through the flow, optimizing for the source (noise) point. We motivate this framework by our key observation stating that for Diffusion/FM models trained with Gaussian probability paths, differentiating through the generation process projects gradient on the data manifold, implicitly injecting the prior into the optimization process. We validate our framework on linear and non-linear controlled generation problems including: image and audio inverse problems and conditional molecule generation reaching state of the art performance across all. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_14017 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | D-Flow: Differentiating through Flows for Controlled Generation Ben-Hamu, Heli Puny, Omri Gat, Itai Karrer, Brian Singer, Uriel Lipman, Yaron Machine Learning Taming the generation outcome of state of the art Diffusion and Flow-Matching (FM) models without having to re-train a task-specific model unlocks a powerful tool for solving inverse problems, conditional generation, and controlled generation in general. In this work we introduce D-Flow, a simple framework for controlling the generation process by differentiating through the flow, optimizing for the source (noise) point. We motivate this framework by our key observation stating that for Diffusion/FM models trained with Gaussian probability paths, differentiating through the generation process projects gradient on the data manifold, implicitly injecting the prior into the optimization process. We validate our framework on linear and non-linear controlled generation problems including: image and audio inverse problems and conditional molecule generation reaching state of the art performance across all. |
| title | D-Flow: Differentiating through Flows for Controlled Generation |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2402.14017 |