{"id":"arxiv/2602.13620","url":"https://arxiv.org/abs/2602.13620","tags":["math.OC","math.NA"],"score":0,"title":"An adaptive framework for first-order gradient methods","domain":"math.OC","source":"arxiv","authors":["Xiaozhe Hu","Sara Pollock","Zhongqin Xue","Yunrong Zhu"],"reviews":0,"subject":"math.OC","summary":"Gradient methods are widely used in optimization problems. In practice, while the smoothness parameter can be estimated utilizing techniques such as backtracking, estimating the strong convexity parameter remains a challenge; moreover, even with the optimal parameter choice, convergence can be slow. In this work, we propose a framework for dynamically adapting the step size and momentum parameters in first-order gradient methods for the optimization problem, without prior knowledge of the strong convexity parameter. The main idea is to use the geometric average of the ratios of successive…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-02-14","reactions":{}}