Homomorphic Encryption And Its Application In Privacy Preserving Machine Learning

With the rapid development of machine learning technology and its applications in all fields of artificial intelligence,machine learning data continues to grow,and industry personnel and the public are increasingly concerned about data privacy and security.Machine learning algorithms encode the relationships on the data by analyzing large amounts of data and updating its model parameters.Due to data sources,the computing nodes and multiple participants,the current machine systems are faced with serious privacy leakage problems.With the rapid development of quantum computing,the research on information security under the environment of quantum computing is also extra urgent.To realize the machine learning protocol of anti-quantum attack privacy protection is an urgent requirement to guarantee network security and user privacy protection.Therefore,privacy protection in machine learning is the focus of current and future research in the field of information security.The research is focus on following topics:1.We consider securely outsourced face recognition under federated cloud environment.We propose a novel and efficient scheme to the PPOFR problem with outsourced computation for the first time under a federated cloud environment based on the Eigenfaces algorithm.We theoretically estimate both the computation and communication costs of the proposed protocol,which efficiently protects data confidentiality of the participating entities under the standard semi-honest model.The proposed protocol can protect the privacy and reduce the computation cost of the client and the face database server.2.We consider privacy of outsourced two-party K-means clustering.The challenges of clustering on two-party encrypted data are analyzed.Outsourced scheme of clustering two-party encryption data is constructed based on the comparison of the two encryption distances.we demonstrate that,by using homomorphic encryption,it is possible to outsource the execution of a two-party k-means clustering algorithm to a single cloud server while retaining confidentiality of the test data.The obtained results improve and generalize the results in [46,56].3.We consider two-party privacy preserving set intersection with FHE.We present a novel private set intersection protocol based on compressed fully homomorphic encryption scheme SGFHE and prove the security of the protocol in the semi-honest model.We also present a variant of the protocol which is a completely novel construction for computing the intersection based on Bloom filter and fully homomorphic encryption.Its efficiency is independent of the size of the customer set.We also present outsourced two-party private set intersection protocols.In the cloud computing environment,it is easy to extend the protocols to on malicious server.4.We consider multi-party computation protocol based on compressed multikey homomorphism fully homomorphic encryption(MKFHE).The homomorphism of SGFHE key is analyzed,and a threshold decryption MKFHE protocol is constructed under the semi-honest,semi-malicious and common random string models.Its security relies on the LWE and RLWE problems(binary secret keys).At last,we summarize the main results of this thesis,and introduce our followup research work and future prospects.