{"id":"arxiv/2603.29276","url":"https://arxiv.org/abs/2603.29276","tags":["math.OC"],"score":0,"title":"Adaptive Momentum via Minimal Dual Function for Accelerating Randomized Sparse Kaczmarz","domain":"math.OC","source":"arxiv","authors":["Lu Zhang","Jinchuan Zeng","Hongxia Wang","Hui Zhang"],"reviews":0,"subject":"math.OC","summary":"Recently, the randomized sparse Kaczmarz method has been accelerated by designing heavy ball momentum adaptively via a minimal-error principle. In this paper, we develop a new adaptive momentum method based on the minimal dual function principle to go beyond the exact measurement restriction of the minimal-error principle. Moreover, by integrating the new adaptive momentum method with the quantile-based sampling, we introduce a general algorithmic framework, called quantile-based randomized sparse Kaczmarz with minimal dual function momentum, which provides a unified approach to exact,…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-03-31","reactions":{}}