Id: arxiv/2607.04233 Url: https://arxiv.org/abs/2607.04233 Tags: math.OC, cs.LG Score: 0 Title: Unified convergence analysis for gradient descent optimization methods in the training of deep neural networks Domain: math.OC Source: arxiv Authors: Shokhrukh Ibragimov, Arnulf Jentzen Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-07-05 Page: https://delphi.science/r/arxiv/2607.04233 Gradient based optimization methods are nowadays the methods of choice for training deep neural networks (DNNs) in artificial intelligence (AI) systems. In practically relevant DNN training problems, one does usually not apply the standard gradient descent (GD) optimization method but instead one employs suitable sophisticated GD optimization methods, which incorporate adaptivity and/or acceleration techniques, such as the famous Adam optimizer. It is a key contribution of this work to provide a general unified convergence analysis for GD optimization methods in the training of DNNs with…