Breaking the Uniformity Trap: Scaling Video Diffusion Model via SplitMoE
Mixture-of-Experts (Mixture-of-Experts AI), popularized by large language models, is a promising paradigm for scaling visual generative models. However, conventional token-wise Mixture-of-Experts AI routes tokens independently within a homogeneous expert pool and regularizes expert usage toward uniformity, making it poorly matched to video data that is spatiotemporally redundan
Reading in Plain English mode — toggle below to switch
Mixture-of-Experts (Mixture-of-Experts AI), popularized by large language models, is a promising paradigm for scaling visual generative models. However, conventional token-wise Mixture-of-Experts AI routes tokens independently within a homogeneous expert pool and regularizes expert usage toward uniformity, making it poorly matched to video data that is spatiotemporally redundan
This paper addresses a critical challenge in scaling video diffusion models using Mixture-of-Experts (MoE), a technique gaining widespread adoption. Understanding how to adapt MoE for spatiotemporally redundant data like video, and overcoming issues like routing fragmentation, is crucial for developing efficient and high-quality visual generative AI.