{"id":"arxiv/2610.14764","url":"https://arxiv.org/abs/2610.14764","tags":["math.OC","math.NA"],"score":0,"title":"A Unified Framework for Randomized Row-Action Methods for Structured Data Fitting and Constraint Geometry","domain":"math.OC","source":"arxiv","authors":["Valentin Leplat","Zeyu Dong","Junfeng Yin"],"reviews":0,"subject":"math.OC","summary":"We study randomized row-action methods for problems with linear observations, convex constraints, and non-Euclidean geometry. Our starting point is a unified formulation in which the data-fitting term acts in observation space, while the feasible set and the regularizer encode structure in variable space. It covers constrained least squares, nonnegative and simplex-constrained models, regularized data fitting, and equality-constrained Bregman selection problems. The structure of these problems leads naturally to two algorithmic branches. The first uses primal Bregman-proximal row steps for…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-10","reactions":{}}