arXiv stat.ML
· Papers
Local Fokker–Planck Geometry for Score Estimation: Heat-Ball Mean-Value Representations and Exact High-Dimensional Sampling
arXiv:2606.27954v1 Announce Type: new Abstract: Score-based generative models and Langevin samplers rely on estimating the score function $nabla_xlog p_t(x)$ of a forward diffusion. Classically this is tractable when the drift is linear: the marginal density is Gaussian and the score is a global conditional expectati