{"id":"arxiv/2608.12009","url":"https://arxiv.org/abs/2608.12009","tags":["math.OC","cs.LG"],"score":0,"title":"Adaptive Bregman Proximal Stochastic Gradient with a Stabilized Barzilai--Borwein Step Size","domain":"math.OC","source":"arxiv","authors":["Chenhan Jin","Shengze Xu","Binghui Xie","Kaiwen Zhou","Fan Jia","James Cheng","Tieyong Zeng"],"reviews":0,"subject":"math.OC","summary":"Bregman proximal stochastic gradient (BPSG) methods bring variance-reduced composite optimization to objectives whose geometry is poorly captured by Euclidean smoothness. Their performance, however, remains sensitive to the step size: raw stochastic curvature estimates can fluctuate sharply, whereas line searches add repeated proximal evaluations. We introduce Ada-BPSG, a line-search-free BPSG method that couples the SAGA gradient table with a stabilized Barzilai--Borwein (BB) candidate. A mediant aggregates incremental secant information so that nearly singular local ratios receive little…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-12","reactions":{}}