{"id":"arxiv/2610.38197","url":"https://arxiv.org/abs/2610.38197","tags":["cs.LG","cs.AI","math.OC"],"score":0,"title":"Convex-SKM-Net: Low-Latency Block Halfspace Corrections for Neural Convex Constraint Satisfaction","domain":"cs.LG","source":"arxiv","authors":["Haoyu Zhu","Xiaochen Zhang","Qingchun Hou"],"reviews":0,"subject":"cs.LG","summary":"Fast inference and strict constraint satisfaction are essential for learning-based decision-making. Existing methods often rely on constraint-specific projections or feasible-set parameterizations, while iterative repair can be costly. We introduce Convex-SKM-Net, a convex constraint-satisfaction layer that forms linearization halfspaces from residual and (sub)gradient oracles without constraint-specific inequality projectors. It applies fixed-depth dual projected-gradient descent (dual PGD) to minimum-displacement block problems using GPU-friendly matrix operations. We jointly learn the…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-10-28","reactions":{}}