Alan Hou

Latest article / 最新文章

Multi-View MoE: Letting Every Router Shape Expert Computation

Multi-View MoE：让每一个路由器协同塑造专家计算与表征

Traditional Sparse Mixture-of-Experts (MoE) models suffer from a fundamental disconnect: single-router selection scores correlate poorly with true empirical expert utility in deep layers. Multi-View MoE resolves this without extra compute or communication overhead by aggregating independent multi-routers and reusing routing preferences to modulate latent-view representations before expert dispatch.