Cyclic denoising drives diffusion model samples to ultrastable attractors that often match memorized training images, including watermarks and artifacts, across multiple model types.
Random organization in periodically driven systems
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Cyclic Denoising Reveals Ultrastable Memories in Diffusion Models
Cyclic denoising drives diffusion model samples to ultrastable attractors that often match memorized training images, including watermarks and artifacts, across multiple model types.