AI RESEARCH
Understanding Latent Diffusability via Fisher Geometry
arXiv CS.LG
•
ArXi:2604.02751v1 Announce Type: new Diffusion models often degrade when trained in latent spaces (e.g., VAEs), yet the formal causes remain poorly understood. We quantify latent-space diffusability through the rate of change of the Minimum Mean Squared Error (MMSE) along the diffusion trajectory. Our framework decomposes this MMSE rate into contributions from Fisher Information (FI) and Fisher Information Rate (FIR). We nstrate that while global isometry ensures FI alignment, FIR is governed by the encoder's local geometric properties.