Design And Development Of Secure Multi-party Computing Platform For Machine Learning

With the arrival of the big data era,data has become the most basic production factor.How to securely exchange and process data has become an urgent issue for the data industry to solve.Secure multi-party computing technology for data security and privacy protection,which can use the private data of participants to complete data processing without disclosing additional information.Therefore,studying the use of secure multi-party computing technology to protect data privacy in machine learning is an important topic for data security issues in artificial intelligence data processing.This paper designs and develops a secure multi-party computing platform for machine learning,which can support machine learning algorithms such as decision tree and support vector machine.The platform consists of modules such as function compiler,circuit generation,and secure computing.The specific content includes:(1)A function compiler has been designed to compile to compile function strings based on regular expressions and syntax tree parsing into protocol code that can run on secure multiparty computing platforms.Using this function compiler to generate code can reduce the impact of human factors on the code.Compared with traditional programming methods,it can avoid commonly used programming errors and vulnerabilities,greatly reducing the time and cost of development and debugging.(2)Designed a secure two-party support vector machine protocol.The protocol can use the horizontal distribution training data of two participants to calculate the support vector machine to get the hyperplane equation.This function adopts a separate design method,which can replace different kernel functions according to actual situations.Compared to other secure multi-party support vector machine protocols,the protocol designed in this paper has better scalability and applicability.This article conducted benchmark testing and accuracy testing on the implementation of the protocol on the secure multi-party computing platform,proving its feasibility and effectiveness.(3)Designed a secure two-party decision tree protocol.This protocol uses the horizontally distributed training data provided by the participants to jointly train and generate a decision tree,and adds a security pre-pruning mechanism,which improves the prediction accuracy.This article conducted benchmark testing and accuracy testing on the implementation of this protocol on a secure multi-party computing platform,and the results showed that its accuracy improved by about 5%compared to other uncut secure decision tree protocols.