#artificial intelligence
Jun 2026
Wasserstein Convergence of ODE-Based Samplers in Decentralized Diffusion Model via Velocity Field Decomposition
A convergence guarantee in Wasserstein-2 distance is established, showing that the distribution of the $N$-step discretization converges to the analytical solution at rate $\mathcal{O}(N^{-1/2}+\varepsilon)$ in $W_2$, where $\varepsilon$ captures the neural approximation errors.
Chenchen Tang, Xuanyu Xue, Fangyikang Wang et al.
· arXiv.org · 0 citations