Privacy-Preservation For Online Distributed Dual Averaging Algorithm

In recent years,with the popularization of the Internet,information industrialization technology has developed rapidly.The breakthrough progress of data storage and data mining technology has made distributed systems based on the Internet of Things-Big Data-Cloud computing a trend of development.Optimization is important for the design,deployment,and function of many distributed systems.In distributed optimization,individuals communicate and coordinate with each other through the network to accomplish computing tasks cooperatively.In most distributed optimization algorithms,this kind of information interaction is explicit.When the interactive information contains individual sensitive data,it is easy to cause privacy disclosure.Moreover,because the environment is full of uncertainty,the individual cost functions and network topology are easy to change.To solve the above two problems,this paper proposes the distributed dual average online algorithm(ODDA-PP)with privacy protection and the switching network average consensus algorithm with privacy protection.Firstly,a weight decomposition mechanism is proposed for the existing distributed dual average online algorithm(ODD).By reasonably decomposing the weight into weight pairs and combining Paillier Cryptosystem,the privacy protection is realized without the assistance of the third party by utilizing the property of addition homomorphism.The theoretical derivation and numerical simulation of the ODDA-PP algorithm prove that the Regret bound is O((?)).Secondly,to solve the problem of privacy protection in switching networks,an average consensus algorithm with privacy protection is proposed.Through the state decomposition mechanism,the true state of the individual is decomposed into explicit state and implicit state.By using the improved update rule,not only the true state of the individual is protected by privacy,but also the convergence analysis and numerical simulation show that all the individuals can converge to the accurate average consensus without losing the accuracy.The two algorithms designed in this paper can not only achieve privacy protection without affecting the optimality of the solution,but also realize the real-time processing of data in distributed optimization and enhance the data security of distributed systems.The theory and simulation prove that the two algorithms have high efficiency and low computational complexity.Analysis shows that the weight decomposition and state decomposition mechanisms do not affect the convergence of the algorithm,so the implementation efficiency of the algorithm can be improved by designing the network topology.Figure[27]Table[0]Reference[48]...