Improving Consistency Models with Generator-Augmented Flows
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916821768077312 |
|---|---|
| author | Issenhuth, Thibaut Lee, Sangchul Santos, Ludovic Dos Franceschi, Jean-Yves Kim, Chansoo Rakotomamonjy, Alain |
| author_facet | Issenhuth, Thibaut Lee, Sangchul Santos, Ludovic Dos Franceschi, Jean-Yves Kim, Chansoo Rakotomamonjy, Alain |
| contents | Consistency models imitate the multi-step sampling of score-based diffusion in a single forward pass of a neural network. They can be learned in two ways: consistency distillation and consistency training. The former relies on the true velocity field of the corresponding differential equation, approximated by a pre-trained neural network. In contrast, the latter uses a single-sample Monte Carlo estimate of this velocity field. The related estimation error induces a discrepancy between consistency distillation and training that, we show, still holds in the continuous-time limit. To alleviate this issue, we propose a novel flow that transports noisy data towards their corresponding outputs derived from a consistency model. We prove that this flow reduces the previously identified discrepancy and the noise-data transport cost. Consequently, our method not only accelerates consistency training convergence but also enhances its overall performance. The code is available at: https://github.com/thibautissenhuth/consistency_GC. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_09570 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Improving Consistency Models with Generator-Augmented Flows Issenhuth, Thibaut Lee, Sangchul Santos, Ludovic Dos Franceschi, Jean-Yves Kim, Chansoo Rakotomamonjy, Alain Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition Consistency models imitate the multi-step sampling of score-based diffusion in a single forward pass of a neural network. They can be learned in two ways: consistency distillation and consistency training. The former relies on the true velocity field of the corresponding differential equation, approximated by a pre-trained neural network. In contrast, the latter uses a single-sample Monte Carlo estimate of this velocity field. The related estimation error induces a discrepancy between consistency distillation and training that, we show, still holds in the continuous-time limit. To alleviate this issue, we propose a novel flow that transports noisy data towards their corresponding outputs derived from a consistency model. We prove that this flow reduces the previously identified discrepancy and the noise-data transport cost. Consequently, our method not only accelerates consistency training convergence but also enhances its overall performance. The code is available at: https://github.com/thibautissenhuth/consistency_GC. |
| title | Improving Consistency Models with Generator-Augmented Flows |
| topic | Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2406.09570 |