Snap-Shot Decentralized Stochastic Gradient Tracking Methods
From MaRDI portal
(Redirected from Publication:6420129)
Abstract: In decentralized optimization, agents form a network and only communicate with their neighbors, which gives advantages in data ownership, privacy, and scalability. At the same time, decentralized stochastic gradient descent ( exttt{SGD}) methods, as popular decentralized algorithms for training large-scale machine learning models, have shown their superiority over centralized counterparts. Distributed stochastic gradient tracking~( exttt{DSGT})~citep{pu2021distributed} has been recognized as the popular and state-of-the-art decentralized exttt{SGD} method due to its proper theoretical guarantees. However, the theoretical analysis of dsgt~citep{koloskova2021improved} shows that its iteration complexity is , where is a double stochastic mixing matrix that presents the network topology and is a parameter that depends on . Thus, it indicates that the convergence property of exttt{DSGT} is heavily affected by the topology of the communication network. To overcome the weakness of exttt{DSGT}, we resort to the snap-shot gradient tracking skill and propose two novel algorithms. We further justify that the proposed two algorithms are more robust to the topology of communication networks under similar algorithmic structures and the same communication strategy to dsgt~. Compared with dsgt, their iteration complexity are and which reduce the impact on network topology (no ).
This page was built for publication: Snap-Shot Decentralized Stochastic Gradient Tracking Methods
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6420129)
