Feature Selective Model Collapse in Diffusion Models: Total Replacement versus Fixed-Budget Training

Excerpt

arXiv:2610.01318v1 Announce Type: new Abstract: Model collapse arises when generative models are trained on synthetic data produced by earlier models. The phenomenon has attracted considerable attention because of its societal and technical implications. However, previous studies have reached seemingly contradictory conclusions: replacing real data with synthetic data causes collapse (Shumailov et al.), yet accumulating real data alongside synthetic data can prevent it. For diffusion models, we