Prediction--Loss Alignment for Sampler--Robust Flow Matching Training
This work investigates the integrability of the pre-optimizer stochastic-gradient second moment through the integrability of the pre-optimizer stochastic-gradient second moment of prediction--loss alignment and shows that prediction--loss alignment eliminates this conversion-induced source of non-integrability.
Jia-Dong Hong, Lei Liu, Xinyu Bian et al.
· 2 citations