{"id":"arxiv/2601.01493","url":"https://arxiv.org/abs/2601.01493","tags":["cs.LG","math.OC"],"score":0,"title":"Accelerating Decentralized Optimization via Overlapping Local Steps","domain":"cs.LG","source":"arxiv","authors":["Yijie Zhou","Shi Pu"],"reviews":0,"subject":"cs.LG","summary":"Decentralized optimization has emerged as a critical paradigm for distributed learning, enabling scalable training while preserving data privacy through peer-to-peer collaboration. However, existing methods often suffer from communication bottlenecks due to frequent synchronization between nodes. We present Overlapping Local Decentralized SGD (OLDSGD), a novel approach to accelerate decentralized training by computation-communication overlapping, significantly reducing network idle time. With a deliberately designed update, OLDSGD preserves the same average update as Local SGD while avoiding…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-01-04","reactions":{}}