Research
Geometric Decoupling: Diagnosing the Structural Instability of Latent
arXiv:2604.18804v1 Announce Type: cross Abstract: Latent Diffusion Models (LDMs) achieve high-fidelity synthesis but suffer from latent space brittleness, causing discontinuous semantic jumps during e
arXiv:2604.18804v1 Announce Type: cross Abstract: Latent Diffusion Models (LDMs) achieve high-fidelity synthesis but suffer from latent space brittleness, causing discontinuous semantic jumps during editing. We introduce a Riemannian framework to diagnose this instability by analyzing the generative Jacobian, decomposing geometry into extit{Local Scaling} (capacity) and extit{Local Complexity} (curvature). Our study uncovers a extbf{Geometric Decoupling"}: while curvature in normal generation functionally encodes image detail, OOD generation exhibits a functional decoupling where extreme curvature is wasted on unstable semantic boundaries rather than perceptible details. This geometric misallocation identifies Geometric Hotspots" as the structural root of instability, providing a robust intrinsic metric for diagnosing generative reliability.
Related
- CLIMB: Controllable Longitudinal Brain Image Generation using Mamba-based Latent Diffusion Model and Gaussian-aligned Autoencoder
- ChopGrad: Pixel-Wise Losses for Latent Video Diffusion via Truncated Backpropagation
- Noise Aggregation Analysis Driven by Small-Noise Injection: Efficient Membership Inference for Diffusion Models
- DeCo: Frequency-Decoupled Pixel Diffusion for End-to-End Image Generation
- From Navigation to Refinement: Revealing the Two-Stage Nature of Flow-based Diffusion Models through Oracle Velocity
Source: arXiv cs.AI | 2026-04-22