Distributed optimization for intelligent IoT under unstable communication conditions

Abstract

The design of distributed optimization algorithms for intelligent Internet of Thing (IIoT) environments has attracted extensive attention. However, the dynamic network environment, the delay of information transmission between devices and the risk of privacy disclosure pose great challenges to algorithm designs. To address this problem, we propose a distributed gradient descent algorithm, which can converge under very mild assumptions that there will be an unbounded (stochastic) delay from sender to receiver. To the best of our knowledge, our proposed algorithm is the first work to take unbounded stochastic communication delays into consideration, in addition to time-varying networks and privacy disclosure. Rigorous analysis shows that our algorithm can converge under such mild assumptions. Extensive experiments exhibit that the proposed algorithm performs well in complex IIoT environments.