| With the rapid development of information technology,massive high-dimensional sensitive data have been collected in various fields.These data are usually stored in a distributed manner,how to extract useful information from these data and protect personal privacy has become a hot spot in many fields,such as statistics,information science and computer science.Existing distribut-ed algorithms do not have the ability to protect personal privacy.Therefore,it is of great significance to carry out the research of distributed algorithms with privacy protection ability.This thesis focuses on distributed algorithms based on differential privacy,specifically:First,we focus on the protection of group structure data privacy in dis-tributed storage,and carry out research on distributed group variable selection that satisfies differential privacy.Firstly,based on the Alternating Direction Method of Multiplier(ADMM),a distributed Logistic group variable selection algorithm is proposed.Furthermore,in order to prevent possible privacy leak-age in the process of computer information interaction,a distributed Logistic randomized response group variable selection algorithm is proposed,and it is proved that the algorithm satisfies differential privacy.Experiments show that the proposed algorithm can effectively deal with group structure classification data and protect its privacy.Second,under given assumptions,the research of distributed online learn-ing based on differential privacy is carried out.Firstly,a distributed online logistic model is proposed,and then the model is solved based on the Online Alternating Direction Method of Multiplier(OADMM).Aiming at the problem of possible privacy leakage,a distributed online Logistic gradient perturbation algorithm is proposed.It is proved that the proposed algorithm satisfies differ-ential privacy and gives the Regret bound.The experimental results show that the proposed differential privacy algorithm can effectively deal with distributed storage streaming classification data and provide privacy protection. |
