一种改进的分布式优化随机差分隐私算法

An improved distributed optimized random differential privacy algorithm

多智能体系统中的分布式优化隐私保护,利用邻居的信息来协同最小化所有智能体目标函数的和,旨在潜在窃听者能够获取所有通信信息的条件下,保护每个智能体的目标函数不被窃取。为了进一步提升性能,在梯度跟踪技术与现有DiaDSP算法的基础上,提出了在状态和方向上添加随机噪声的随机差分隐私算法,并证明了该算法不仅能够在概率意义下收敛到问题的最优解,还能够在信息交互过程中保持DiaDSP算法的隐私保护性能,而且通过添加随机的随机变量能够节省计算时间,提高优化效率。数值仿真结果显示,相较于DiaDSP算法,改进算法在p<1时收敛时间减少,收敛精度也有较大的提升。

For the distributed optimization privacy protection problem in multi-agent systems,each agent uses the information of its neighbors to minimize the sum of the objective functions of all agents.The goal is to protect the objective function of each agent under the assumption that potential eavesdroppers have access to all communications.Based on the gradient tracking technology and the existing DiaDSP algorithm,a stochastic differential privacy algorithm with random noise added to the state and direction is proposed,and its convergence and privacy-preserving performance are proved.It can maintain the privacy protection performance of DiaDSP algorithm in the information exchanging process,which can also reduce the computational cost and improve the effectiveness of the algorithm through the randomly added random variables.The algorithm is simulated by an example,the results show that compared with the DiaDSP algorithm,the convergence time is reduced at p<1,and the convergence accuracy is also greatly improved.