Diffusion models generate novel samples due to the interaction between denoiser architecture inductive bias and target distribution, with explicit generated distributions derived for linear, polynomial, and bottleneck architectures.
Generalization in diffusion models arises from geometry-adaptive harmonic representations
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Diffusion Models, Denoiser Architecture and Creativity
Diffusion models generate novel samples due to the interaction between denoiser architecture inductive bias and target distribution, with explicit generated distributions derived for linear, polynomial, and bottleneck architectures.