{"id":"arxiv/2606.02176","url":"https://arxiv.org/abs/2606.02176","tags":["math.OC"],"score":0,"title":"Relax and Follow: L0-Path Computation with L0-Bregman Relaxations","domain":"math.OC","source":"arxiv","authors":["Mhamed Essafri","Luca Calatroni","Emmanuel Soubies"],"reviews":0,"subject":"math.OC","summary":"This work introduces L0PathBrex, a novel method for estimating the solution path of L0-regularized problems through the use of L0 Bregman relaxations (B-rex). Recently introduced and analyzed in the literature, these relaxations provide continuous reformulations of the original objective, are applicable to possibly non-quadratic data fidelity terms, and depend on a family of functions designed to preserve the global minimizers while eliminating part of the undesirable local minima. Given any numerical solver for the relaxation, the proposed approach dynamically constructs a collection of…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-06-01","reactions":{}}