Id: arxiv/2601.21355 Url: https://arxiv.org/abs/2601.21355 Tags: math.OC Score: 0 Title: Decentralized Learning with Dynamically Refined Edge Weights: A Data-Dependent Framework Domain: math.OC Source: arxiv Authors: Rongxing Du, Hoi-To Wai Reviews: 0 Subject: math.OC Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-01-29 Page: https://delphi.science/r/arxiv/2601.21355 This paper aims to accelerate decentralized optimization by strategically designing the edge weights used in the agent-to-agent message exchanges. We propose a Dynamic Directed Decentralized Gradient (D3GD) framework and show that the proposed data-dependent framework is a practical alternative to the classical directed DGD (Di-DGD) algorithm for learning on directed graphs. To obtain a strategy for edge weights refinement, we derive a design function inspired by the cost-to-go function in a new convergence analysis for Di-DGD. This results in a data-dependent dynamical design for the edge…