Researches On Privacy-preserving Combinatorial Optimization And Its Applications For Medical Data

With the development of distributed computing storage technologies,the collaborative combinatorial optimization of data between multiple parties has become more common,and it is applied to many basic fields such as logistics,social networks,and medical fields.Although distributed parties can collect more data to provide better data support for optimization issues,distributed optimizations must provide local information such as location and physiological data.The sharing of confidential information will cause serious privacy issues,which can only be alleviated by limiting data distribution.Therefore,how to protect the privacy and security of users’ data gradually attract the widespread concern of relevant workers while collecting distributed data and performing combinatorial optimization.This paper aims to construct some privacy-preserving combinatorial optimization methods,which enable organizations or individuals to participate in collaborative optimization without leaking local sensitive information.This paper specifically studies and addresses privacy issues in distributed fractional knapsack and shortest path optimization,and gives practical applications.The work is summarized as follows:(1)This paper considers that the optimization parameters(objective function and constraint conditions)are partitioned between two parties,that is,one party privately holds the objective function and the other party privately holds the constraint conditions.For the data privacy issue of two parties in the distributed combinatorial optimization,this paper adopts the Paillier cryptosystem to aggregate and process the distributed data through its homomorphic property.In order to efficiently obtain an effective solution,this paper designs a secure transformation approach to ensure secure input of data,and proposes the secure and complete two-party comparison and sorting protocols to solve privacy issues of fractional knapsack optimization through greedy idea.Through simulation experiments,this paper shows that the solution with privacy protection and the solution under plaintext are consistent,and total cost is approximately one second.(2)In addition to the two-party model,this paper further studies the more complex case that the parameters are arbitrarily to many parties,that is,each party privately owns a part of optimization parameters.Considering the iterative communication issues caused by any two parties under the multi-party model,this paper adopts the Shamir secret sharing techniques to outsource computing to cloud servers for simplifying communication and improving efficiency.Different from the two-party optimization,in order to protect some parameters of each entity,this paper uses a secret sharing method to safely upload data to two cloud servers and designs a binary secure comparison protocol to coordinate the aggregation of parameters.For the efficiency issues,this paper designs a privacy-preserving Min-Heap implementation method to ensure that the private information is not leaked during path query in the weighted directed connected graph.As a result,the privacy-preserving optimization of shortest path is realized under the multi-party model.Through simulation experiments,this paper shows that the solution with privacy protection and the solution under plaintext are consistent.After the data upload and graph construction which only needs to be executed once,the query overhead of shortest path is within one millisecond.This paper combines drug configuration and clinical pathway in medical applications to use the proposed transformation approach and basic protocols for underlying technology.Compared with existing schemes,the proposed scheme has a broader application prospect.By analyzing overhead and the simulation results,this paper proves the feasibility,applicability and scalability of the proposed schemes.