A Method to Improve Consensus Averaging using Quantized ADMM
By Nandan Sriranga, Chandra Ramabhadra Murthy, Vaneet Aggarwal. 2019
9 citations recorded by the source corpus.
In this paper, a method is proposed to overcome the consensus error in an average consensus problem, under a distributed setting and with finite-bit communications between the network agents. Previous works have illustrated that the average consensus problem can be solved under a distributed setting, using the Alternating Direction Method of Multipliers (ADMM) method, in which consensus can be attained by using locally available information along with information from neighboring nodes. This holds true even under finite-bit exchanges between neighbouring nodes, but suffers from consensus errors and cyclic states due to the introduced quantization schemes. This work deals with achieving perfect consensus with finite-bit communications between neighboring nodes. We propose an algorithm which leads to perfect consensus in an asymptotic sense, without the need to increase the per-exchange communication rate of the network.
Subfields: Computer Networks and Communications, Mechanical Engineering
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Networks and Communications
- Jihong Park
- Medhi Bennis
- Anis Elgabli
- Vaneet Aggarwal
- Amrit Singh Bedi
- Merouane Debbah
- Sumudu Samarakoon
- Walid Saad
In Mechanical Engineering
Citations in the corpus
- GADMM: Fast and Communication Efficient Framework for Distributed Machine Learning (2019, background): “For a large scale network, the communication cost often becomes dominant compared to the computation time, calling for communication efficient distributed optimization (Zhang et al., 2012; McMahan et al., 2017; Park et al., 2018; Jordan et al., 2018; Liu et al., 2019; Sriranga et al., 2019).”
- Q-GADMM: Quantized Group ADMM for Communication Efficient Decentralized Machine Learning (2019, unknown): “For a large scale network, the communication cost often becomes dominant compared to the computation time, calling for communication efficient distributed optimization [5], [39]–[41].”