Gradient Flow Drifting: Generative Modeling via Wasserstein Gradient Flows of KDE-Approximated Divergences

Researchers introduce Gradient Flow Drifting, a new mathematical framework for generative AI models that connects the Drifting Model to Wasserstein gradient flows of KL divergence under kernel density estimation. The framework includes a mixed-divergence strategy to avoid mode collapse and extends to Riemannian manifolds for improved semantic space applications.

• →New generative AI model family called Gradient Flow Drifting provides precise mathematical framework for improved model generation.
• →Researchers prove equivalence between Drifting Model and Wasserstein gradient flow of forward KL divergence under KDE approximation.
• →Framework includes MMD-based generators as special cases of Wasserstein gradient flows with different divergences.
• →Mixed-divergence strategy combining reverse KL and chi-squared divergence prevents both mode collapse and mode blurring.
• →Extension to Riemannian manifolds reduces kernel function constraints and improves semantic space suitability.