.png){kind=icon}
返回
Fast Sampling of Score-Based Models With Cyclical Diffusion Sampling
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2024.3365146.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Diffusion models have recently exhibited significant potential in generative modeling, surpassing generative adversarial networks concerning perceptual quality and autoregressive models in density estimation. However, a notable drawback of these models is their slow sampling time, requiring numerous model evaluations to generate high-quality samples. This research proposes a technique termed cyclical diffusion sampling that incorporates cyclical stochastic gradient Langevin dynamics (SGLD) with elucidated diffusion models (EDM) sampler to enhance stability when utilizing a limited number of sampling steps. Cyclical step-size scheduling has been demonstrated to enhance the effectiveness of SGLD in learning complex multimodal distributions. We have shown, via empirical assessments, that cyclical diffusion sampling significantly enhances image quality and markedly decreases inference time. Notably, the method preserves simplicity and requires no alterations to the network architecture, which promotes straightforward reproducibility and seamless integration with current methodologies.
Keyword:
Mathematical models
Computational modeling
Solid modeling
Probabilistic logic
Noise reduction
Training
Perturbation methods
Stochastic processes
Score-based generative models
denoising diffusion probabilistic models
Langevin dynamics
stochastic differential equation
期刊
IF:
3.6
论文数:
9.8W
被引数:
29.4W
机构
引用论文
Clinical outcomes of closed-loop vagal nerve stimulation in patients with refractory epilepsy.png)
Seizure
OAAI
SeizureIF0
OAAI分享收藏S100β Interaction with Tau Is Promoted by Zinc and Inhibited by Hyperphosphorylation in Alzheimer's Disease在阿尔茨海默氏病中，锌促进S100β 与Tau的相互作用，并被过度磷酸化抑制。OAAI
OAAI分享收藏Effect of irregular heat source/sink on the radiative thin film flow of MHD hybrid ferrofluid不规则热源/热沉对MHD混合铁磁流体辐射薄膜流动的影响
PREAI
分享收藏 没有更多内容
没有更多内容