Analyzing and Improving the Training Dynamics of Diffusion Models
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2023
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866916167798489088 |
|---|---|
| author | Karras, Tero Aittala, Miika Lehtinen, Jaakko Hellsten, Janne Aila, Timo Laine, Samuli |
| author_facet | Karras, Tero Aittala, Miika Lehtinen, Jaakko Hellsten, Janne Aila, Timo Laine, Samuli |
| contents | Diffusion models currently dominate the field of data-driven image synthesis with their unparalleled scaling to large datasets. In this paper, we identify and rectify several causes for uneven and ineffective training in the popular ADM diffusion model architecture, without altering its high-level structure. Observing uncontrolled magnitude changes and imbalances in both the network activations and weights over the course of training, we redesign the network layers to preserve activation, weight, and update magnitudes on expectation. We find that systematic application of this philosophy eliminates the observed drifts and imbalances, resulting in considerably better networks at equal computational complexity. Our modifications improve the previous record FID of 2.41 in ImageNet-512 synthesis to 1.81, achieved using fast deterministic sampling.
As an independent contribution, we present a method for setting the exponential moving average (EMA) parameters post-hoc, i.e., after completing the training run. This allows precise tuning of EMA length without the cost of performing several training runs, and reveals its surprising interactions with network architecture, training time, and guidance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2312_02696 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Analyzing and Improving the Training Dynamics of Diffusion Models Karras, Tero Aittala, Miika Lehtinen, Jaakko Hellsten, Janne Aila, Timo Laine, Samuli Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning Neural and Evolutionary Computing Diffusion models currently dominate the field of data-driven image synthesis with their unparalleled scaling to large datasets. In this paper, we identify and rectify several causes for uneven and ineffective training in the popular ADM diffusion model architecture, without altering its high-level structure. Observing uncontrolled magnitude changes and imbalances in both the network activations and weights over the course of training, we redesign the network layers to preserve activation, weight, and update magnitudes on expectation. We find that systematic application of this philosophy eliminates the observed drifts and imbalances, resulting in considerably better networks at equal computational complexity. Our modifications improve the previous record FID of 2.41 in ImageNet-512 synthesis to 1.81, achieved using fast deterministic sampling. As an independent contribution, we present a method for setting the exponential moving average (EMA) parameters post-hoc, i.e., after completing the training run. This allows precise tuning of EMA length without the cost of performing several training runs, and reveals its surprising interactions with network architecture, training time, and guidance. |
| title | Analyzing and Improving the Training Dynamics of Diffusion Models |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Machine Learning Neural and Evolutionary Computing |
| url | https://arxiv.org/abs/2312.02696 |