Skip to content

Similarity-Driven Flow Matching (SDFlow)

Home / Concepts / Similarity-Driven Flow Matching (SDFlow)

Similarity-Driven Flow Matching (SDFlow)

Auto-generated stub. Edit this file to add more details.

A non-autoregressive time-series generation framework that uses flow matching on frozen VQ latent spaces to eliminate exposure bias.

Why It Matters

Moves time-series generation away from autoregressive VQ-VAE token modeling toward a parallelizable non-autoregressive flow matching paradigm, directly mitigating exposure bias.

Evidence

SDFlow (Similarity-Driven Flow Matching), a non-autoregressive framework that operates entirely in the frozen VQ latent space and enables parallel sequence generation via flow matching.

Metadata & Links