Collaborative Computing Of Privacy-Preserving Logistic Regression Based On Homomorphic Encryption

As a new factor of production,data is increasingly becoming a new power source to improve economic and social productivity.However,the problem of data security has greatly hindered the integration and open sharing of data resources.How to use data resources while protecting data privacy is an urgent problem to be solved.Machine learning with privacy protection can complete the training and reasoning of machine learning model without divulging data privacy.It is an effective way to solve the dilemma of data privacy and data utilization.With the rapid development of data sharing and trading market,data analysis on multiple data sets across institutions and even across fields is becoming an urgent demand.However,considering the security of data assets,many data owners are unwilling to transmit data directly to other participants or third parties for calculation.There is an urgent need to design a new multi-party collaborative computing scheme for privacy protection machine learning on multiple data sets.Aiming at the application scenario where the data demander wants to use the data of multiple data owners to calculate the logistic regression model,this thesis proposes a privacy protection logistic regression collaborative calculation method based on homomorphic encryption,which does not depend on the third-party platform,and implements the corresponding prototype system based on the advanced floating-point homomorphic encryption scheme RNS-CKKS.Firstly,a multi-party collaborative computing model composed of data owner,model demander and key generator is designed,which can cooperate to complete the model training task without disclosing the model information and data privacy of all parties.An interactive logic return collaborative computing scheme based on RNS-CKKS cryptosystem is given,and the security of the collaborative computing scheme is analyzed by establishing an attack model.Secondly,the collaborative computing prototype system is implemented on the small computer cluster,and the computing and communication are optimized.In particular,the ciphertext homomorphic operation is unloaded to GPU to improve the computing performance of the system.Finally,an experimental evaluation is carried out on five small medical data sets to verify the correctness of the collaborative computing scheme and the effect of computing optimization measures.The experimental results show that the collaborative computing prototype system can meet the practical requirements on small and medium-sized data sets.