Understanding Latent Diffusability via Fisher Geometry

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Gu, Jing, Mardani, Morteza, Lee, Wonjun, Zou, Dongmian, Lerman, Gilad
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913001264644096
author Gu, Jing
Mardani, Morteza
Lee, Wonjun
Zou, Dongmian
Lerman, Gilad
author_facet Gu, Jing
Mardani, Morteza
Lee, Wonjun
Zou, Dongmian
Lerman, Gilad
contents 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 demonstrate that while global isometry ensures FI alignment, FIR is governed by the encoder's local geometric properties. Our analysis explicitly decouples latent geometric distortion into three measurable penalties: dimensional compression, tangential distortion, and curvature injection. We derive theoretical conditions for FIR preservation across spaces, ensuring maintained diffusability. Experiments across diverse autoencoding architectures validate our framework and establish these efficient FI and FIR metrics as a robust diagnostic suite for identifying and mitigating latent diffusion failure.
format Preprint
id arxiv_https___arxiv_org_abs_2604_02751
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Understanding Latent Diffusability via Fisher Geometry
Gu, Jing
Mardani, Morteza
Lee, Wonjun
Zou, Dongmian
Lerman, Gilad
Machine Learning
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 demonstrate that while global isometry ensures FI alignment, FIR is governed by the encoder's local geometric properties. Our analysis explicitly decouples latent geometric distortion into three measurable penalties: dimensional compression, tangential distortion, and curvature injection. We derive theoretical conditions for FIR preservation across spaces, ensuring maintained diffusability. Experiments across diverse autoencoding architectures validate our framework and establish these efficient FI and FIR metrics as a robust diagnostic suite for identifying and mitigating latent diffusion failure.
title Understanding Latent Diffusability via Fisher Geometry
topic Machine Learning
url https://arxiv.org/abs/2604.02751