Variational Online Mirror Descent for Robust Learning in Schrödinger Bridge
Dong-Sig Han, Jaein Kim, Hee Bin Yoo, Byoung-Tak Zhang · 2025
Read Variational Online Mirror Descent for Robust Learning in Schrödinger Bridge by Dong-Sig Han, Jaein Kim, Hee Bin Yoo, Byoung-Tak Zhang (2025) 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-2504-02618.txt.
Description
Schrödinger bridge methods steer one cloud of probability into another and now power many generative models, but their training signals come with noise that solvers often ignore. Four researchers build a sturdier learning rule in 2025 around online mirror descent, a method that updates step by step as data arrives. They prove it settles down and bound how much it can regret along the way. Their solver, VMSB, needs no simulation and leans on the geometry of Gaussian mixtures.
Across a wide bench of test problems, the new solver beats current bridge methods again and again. The authors credit the variational framing for the steadiness. They offer the work as a bridge solver that stays reliable when the learning signal cannot be trusted.
Catalog record
| CREATOR | Dong-Sig Han, Jaein Kim, Hee Bin Yoo, Byoung-Tak Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 194 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-02618