DDPM (Denoising Diffusion Probabilistic Model)
DDPM (Denoising Diffusion Probabilistic Model): A diffusion model that learns to gradually reverse a noise-addition process. The forward process corrupts data through a Gaussian sequence; the reverse process estimates noise or previous states. Iterative sampling yields high quality but requires many steps.
In italiano: DDPM (Denoising Diffusion Probabilistic Model)Legge II · Probabilità, softmax e modelli generativiThe forward process corrupts data through a Gaussian sequence; the reverse process estimates noise or previous states. Iterative sampling yields high quality but requires many steps.
Examples
- Noise prediction
- Variance schedule
- Iterative sampling