In flow matching, independently pairing noise and data can assign conflicting velocity targets to nearby intermediate states.
Larger optimal transport batches make flow matching more stable
A scalable entropic transport approximation expands online source-target coupling from hundreds to thousands of samples while retaining theoretical and empirical stability benefits.
Top University
Peter Chen · Sergey Levine
University of California, Berkeley
Research Digest··3 min read
Chen and Levine study how flow-matching models pair source noise with target data, a choice that determines the velocity targets used for training.
Why this paper
From University of California, Berkeley
In one line
Larger online entropic OT minibatches reduce gradient variance and integration error in flow matching.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§