基于符号ADD和线性多分支程序的分类算法安全评估

分类算法是机器学习和数据分析中重要的算法.当需要对分类算法本身以及算法的输入数据进行隐私保护时，就出现了分类算法安全评估问题.针对现有的分类算法安全评估协议效率较低的问题，文章给出了一种基于代数决策图和线性多分支程序的解决方案.首先，设计了基于代数决策图的安全函数评估协议，用以安全评估决策函数；其次，引入了线性多分支程序的概念，用其对分类算法进行表示.最后，借助线性多分支程序和基于代数决策图的安全函数评估协议，给出了一个私有线性多分支程序的安全评估协议.对新的协议的正确性和安全性进行了分析和证明.实验数据表明，与原有的解决方案相比，新的协议在效率上有明显的提高.

Secure Evaluation of Classification Algorithms Based on Symbolic ADD and Linear Multi-Branching Program

Classification algorithms are widely used in the areas of machine learning and data mining.It is an important task to evaluate the classification algorithms securely when both the classification algorithm and the input data of the algorithm are private.In order to improve the efficiency of existing secure evaluation protocols,a solution based on both the algebraic decision diagram and the linear multi-branching program was presented.Firstly,a secure function evaluation protocol based on algebraic decision diagram was designed for evaluating decision functions securely.Secondly,a structure named linear multi-branching program was proposed to represent the classification algorithms.Based on both the secure function evaluation protocol and the structure of linear multi-branching program,a protocol for securely evaluating the private linear multi-branching programs was constructed.Both the correctness and the security of the new protocol were analyzed.Experimental results show that the new protocol is more efficient than the existing solutions.