flow

Definition

Names a family of generative models that move probability mass between a simple base distribution and the data distribution through learned continuous transformations — normalizing flows, and more recently flow matching. It sits next to diffusion in the same design space; the difference is that a flow learns a transport path between distributions rather than a denoising chain. Do not confuse it with optical flow, the per-pixel motion field between video frames.

Explainers using this term

• Paper explained: Self-OPD — an image generator that distills itself, with no teacherSelf-OPD: On-Policy Distillation for Flow Matching Models without Teacher
• Flow Matching from Scratch — What Came After Diffusion, and Why It Goes StraightFlow Matching for Generative Modeling
• Paper Deep-Dive: The 'Physics' of Multimodal Pretraining — Which Way Does Knowledge Actually Flow?Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes