Decentralized low-rank matrix completion

Qing Ling Yangyang Xu Wotao Yin Zaiwen Wen

Numerical Linear Algebra mathscidoc:1912.43144

2925-2928, 2012.3
This paper introduces algorithms for the decentralized low-rank matrix completion problem. Assume a low-rank matrix W = [W <sub>1</sub> ,W <sub>2</sub> , ...,W <sub>L</sub> ]. In a network, each agent observes some entries of W <sub></sub> . In order to recover the unobserved entries of W via decentralized computation, we factorize the unknown matrix W as the product of a public matrix X, common to all agents, and a private matrix Y = [Y <sub>1</sub> ,Y <sub>2</sub> , ...,Y <sub>L</sub> ], where Y <sub></sub> is held by agent . Each agent alternatively updates Y <sub></sub> and its local estimate of X while communicating with its neighbors toward a consensus on the estimate. Once this consensus is (nearly) reached throughout the network, each agent recovers W <sub></sub> = XY <sub></sub> , and thus W is recovered. The communication cost is scalable to the number of agents, and W <sub></sub> and Y <sub></sub> are kept private to agent to a certain extent. The algorithm is accelerated by extrapolation and compares favorably to the centralized
No keywords uploaded!
[ Download ] [ 2019-12-21 11:25:32 uploaded by Yangyang_Xu ] [ 193 downloads ] [ 0 comments ]
  title={Decentralized low-rank matrix completion},
  author={Qing Ling, Yangyang Xu, Wotao Yin, and Zaiwen Wen},
Qing Ling, Yangyang Xu, Wotao Yin, and Zaiwen Wen. Decentralized low-rank matrix completion. 2012. pp.2925-2928.
Please log in for comment!
Contact us: | Copyright Reserved