Browse Teach Curate

Conditional Flow Matching for Generative Models

A simulation-free way to train continuous normalizing flows: instead of solving an ODE during training, the model regresses a time-dependent vector field toward a target velocity that transports a simple prior distribution to the data distribution along a conditional probability path. Covers the flow-matching objective, conditional paths (such as optimal-transport or Gaussian interpolants), and how samples are produced by integrating the learned vector field at inference.

jeremy Video

Flow Matching vs Diffusion Models Comparison in Machine Learning Image Generation

Flow Matching represents a deterministic generative modeling framework within machine learning that replaces the stochastic differential equations (SDEs) of diffusion models with ordinary differentia…