A self-consistency property of optimal control yields an iterative online algorithm for learning conditioned diffusion dynamics without trajectory differentiation.
We also discuss potential issues that might arise in certain settings, and present a generalised version of our algorithm which can help to mitigate them
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Control Consistency Losses for Diffusion Bridges
A self-consistency property of optimal control yields an iterative online algorithm for learning conditioned diffusion dynamics without trajectory differentiation.