TY - JOUR
T1 - Decentralized Online Optimization With Compressed Communication Over Directed Graphs
AU - Liu, Honglei
AU - Zhang, Baoyong
AU - Yu, Zhan
AU - Yuan, Deming
AU - Dai, Mingcheng
N1 - This work was supported in part by the National Natural Science Foundation of China under Grant 62273181, Grant 62373190, Grant 62221004, and Grant 12401123; in part by the Research Grants Council of the Hong Kong Special Administrative Region, China, under Grant HKBU 12301424; and in part by the Open Projects of the Institute of Systems Science, Beijing Wuzi University, under Grant BWUISS20
Publisher Copyright:
© 2025 IEEE.
PY - 2025/10/28
Y1 - 2025/10/28
N2 - This article focuses on a decentralized online optimization problem over multiagent systems, where the interactions are modeled by a strongly connected directed graph. The objective of each agent is to minimize the global loss function accumulated by all agents’ local loss functions, which are time-varying and only known by themselves. To address the communication bottleneck caused by the high-dimensional data and large-scale networks, we design a decentralized online algorithm with compressed communication, decentralized online gradient push-sum with compressed communication (CC-DOGPS). For strongly convex functions, a sublinear regret bound O(InT
2) of our designed algorithm is obtained, where T is the time horizon. Finally, two numerical simulations are given to validate the theoretical results and illustrate the efficiency of our designed algorithm.
AB - This article focuses on a decentralized online optimization problem over multiagent systems, where the interactions are modeled by a strongly connected directed graph. The objective of each agent is to minimize the global loss function accumulated by all agents’ local loss functions, which are time-varying and only known by themselves. To address the communication bottleneck caused by the high-dimensional data and large-scale networks, we design a decentralized online algorithm with compressed communication, decentralized online gradient push-sum with compressed communication (CC-DOGPS). For strongly convex functions, a sublinear regret bound O(InT
2) of our designed algorithm is obtained, where T is the time horizon. Finally, two numerical simulations are given to validate the theoretical results and illustrate the efficiency of our designed algorithm.
KW - Compressed communication
KW - decentralized online optimization
KW - directed graph
KW - push-sum
UR - https://www.scopus.com/pages/publications/105020741257
U2 - 10.1109/TNNLS.2025.3622106
DO - 10.1109/TNNLS.2025.3622106
M3 - Journal article
SN - 2162-2388
JO - IEEE Transactions on Neural Networks and Learning Systems
JF - IEEE Transactions on Neural Networks and Learning Systems
ER -
