Time-Reversed BSDEs for Accurate Gradient Estimation in Diffusion Models
Yuhang Mei, Amirhossein Taghvaei · 2026
Read Time-Reversed BSDEs for Accurate Gradient Estimation in Diffusion Models by Yuhang Mei, Amirhossein Taghvaei (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-20455.txt.
Description
Mei and Taghvaei propose a new gradient estimator for fine-tuning diffusion models. Control-based tuning methods often rely on adjoint matching for gradients. That adjoint process is not adapted to the forward diffusion path. The mismatch can yield unstable, high-variance gradient estimates. Their estimator uses a time-reversed backward stochastic differential equation instead. The resulting adjoint stays adapted to the underlying filtration. Analysis and toy experiments show steadier gradients. Performance stayed competitive with adjoint matching in their tests.
Catalog record
| CREATOR | Yuhang Mei, Amirhossein Taghvaei |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37291 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.20455