Conceptmarginal-first
Marginal-First Decomposition
A diffusion model training paradigm that first derives the marginal distribution of the noising process and then constructs the training objective accordingly, allowing reuse of standard Gaussian diffusion training pipelines.
Timeline (last 90 days)
Oct 15
ALD uses a marginal preferential decomposition, expressing stable noise in the form of a scalar Gaussian scale mixture, so that the training objective returns to the standard Gaussian noise prediction loss, requiring only one additional scalar scale input.
Unverified50%