Diffusion Models

Forward Diffusion Process

Training phase where Gaussian noise is progressively added to the original data over several time steps to reach a pure noise distribution. This Markov process is designed to be mathematically reversible and predictable.

← Terug