a new paradigm for generative modeling built on Continuous Normalizing Flows (CNFs)

Flow Matching (FM), a simulation-free approach for training CNFs

1. based on regressing vector fields of fixed conditional probability paths
2. employing FM with diffusion paths results in a more robust and stable alternative for training diffusion models
3. non-diffusion probability paths
   1. using Optimal Transport (OT) displacement interpolation to define the conditional probability paths.
   2. These paths are more efficient than diffusion paths, provide faster training and sampling, and result in better generalization.

FLOW MATCHING FOR GENERATIVE MODELING, CLR2023 spotlight