{"id":"arxiv/2609.27546","url":"https://arxiv.org/abs/2609.27546","tags":["math.ST","cs.LG","math.PR","stat.ML"],"score":0,"title":"Robustness of Diffusion Models under Distribution Shift","domain":"math.ST","source":"arxiv","authors":["Wei Luo","Neil K. Chada","Shijie Zhang","Lu Yu"],"reviews":0,"subject":"math.ST","summary":"Score-based diffusion models are increasingly considered in settings where the underlying data distribution may differ from the training distribution, yet existing theoretical guarantees largely focus on the no-shift setting. In this work, we study robust score estimation under Wasserstein perturbations of a reference distribution. For the Ornstein--Uhlenbeck diffusion, we show that robust estimation decomposes into two fundamental components: the statistical cost of learning the reference distribution and the intrinsic cost of distribution shift. The latter scales quadratically with the…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-23","reactions":{}}