Wasserstein Gradient Flow Analysis of Generative Modeling via Drifting

A new analysis interprets the Generative Modeling via Drifting (GMD) framework through Wasserstein Gradient Flows (WGF), the steepest descent path for a functional in probability measure space under optimal transport geometry. Unlike prior WGF-based work, GMD directly targets a fixed point of a specific WGF. The study shows that one algorithm from Deng et al. (2026) finds the limiting point of a WGF on KL divergence with Parzen smoothing, while the implemented algorithm resembles a fixed point of a WGF on Sinkhorn divergence but lacks certain desirable properties. The note is published on arXiv (2605.05118).