Theoretical Overview

Overview of the generative modeling frameworks implemented in GenSBI, based on the unified theory of score-based diffusion models (Song et al., 2021), the EDM design space (Karras et al., 2022), and conditional flow matching (Lipman et al., 2024).

These pages provide a self-contained introduction to each framework, progressing from the foundational SDE-based score matching to the streamlined EDM formulation, then to flow matching, and finally to the conditional extensions that enable simulation-based inference.

• Score-Based Generative Models
• EDM Diffusion Models
• Flow Matching
• Conditional Density Estimation