The authors build on Energy-Weighted Flow Matching to instantiate endpoint weighting for model-defined ω-powered conditional targets.
Learning powered conditional distributions directly with importance-weighted flow matching
New method reweights training examples to match the ω-powered target, outperforming classifier-free guidance on synthetic and ImageNet data.
Big Tech
Maciej Falkiewicz · Frantzeska Lavda · Jason Ramapuram · Alexandros Kalousis
University of Geneva · HES-SO/HEG Genève · Google Deepmind
Research Digest··2 min read
Falkiewicz et al.
Why this paper
From Google Deepmind and 2 others
In one line
Importance-Weighted Conditional Flow Matching learns an ω-powered conditional distribution more faithfully than classifier-free guidance.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·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.
§