Rational Delegated Learning Model Analysis And Design

With the rapid development of artificial intelligence technologies such as big data,cloud computing,and machine learning,a large amount of beneficial information in various types of data can be mined and utilized.However,due to the complexity of machine learning technology,many companies,institutions,or individuals that do not have data mining capabilities are limited by their own computing capabilities,unable to mine useful information from data.They can only delegate cloud service providers with super-computing ability to mine data,and the resulting security problems such as user privacy data leakage are increasing prominent.Therefore,it can be seen that constructing a secure model outsourcing training scheme to avoid the leakage of user privacy data is essential for achieving data security sharing and mining.This paper uses cryptography technology and game theory as tools,and studies the method of delegating machine learning models training according to traditional delegation computing ideas.The research content involves rational delegation learning schemes based on game theory,joint delegation learning model and protocol based on BCP homomorphic encryption,and application of privacy protection decision tree model in disease prediction.(1)Rational delegation learning schemes based on game theory.Firstly,based on the delegation computing ideas and game theory,a delegation learning model was proposed and a game model was established.Secondly,the rational delegation learning model and scheme were constructed for the decision tree model.Finally,the experimental test results showed that the scheme reduced the client's computing cost,and under the condition of ensuring the security of privacy data,a high-precision decision tree model can be obtained.(2)Joint delegation learning model and protocol based on BCP homomorphic encryption.Firstly,a privacy protection method based on false records is proposed for the structure of the decision tree model.Secondly,the joint delegation learning protocols are designed according to the vertical and horizontal distribution of data.Finally,the results of the security proof and the performance analysis show that the final model obtained by the client is consistent with the construction by real data.(3)Application of privacy protection decision tree model in disease prediction.Firstly,a privacy protection disease prediction model is constructed based on the BCP homomorphic encryption algorithm.Secondly,the privacy protection disease prediction algorithm and protocol are designed for the decision tree model.Finally,the experimental results show that the protocol execution process will not leak privacy data,and there is a satisfactory efficiency when multiple users query at the same time.