Distributionally Robust Optimization via Generative Ambiguity Modeling

Jiaqi Wen, Jianyi Yang · 2026

Read Distributionally Robust Optimization via Generative Ambiguity Modeling by Jiaqi Wen, Jianyi Yang (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-2602-08976.txt.

Description

Machine learning models often stumble when the world shifts away from training data. Jiaqi Wen and one co-author strengthen distributionally robust optimization by building ambiguity sets from generative models. A good ambiguity set must stay faithful to the nominal distribution while covering enough alternative scenarios, and it must keep the optimization tractable.

Their GAS-DRO algorithm solves the inner worst-case search over a parameterized generative space, with proven stationary convergence. Implemented with a diffusion model, it beats rivals on out-of-distribution generalization in machine-learning tasks. The work offers a practical recipe for robustness that reaches beyond the support of the observed data.

Catalog record

CREATORJiaqi Wen, Jianyi Yang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE126 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-08976

Browse the full library · About the Tower of Babel