基于同态加密的分布式加密流量分类隐私保护方法

Privacy Protection Method of Distributed Encrypted Traffic Classification Based on Homomorphic Encryption

随着信息技术的飞速发展,数据量迅速增加,逐渐演变出了分布式存储方式。针对分布式数据存储方式中容易遭受模型训练梯度推理攻击造成梯度泄露,进而引发分布式节点中数据集泄露的问题,提出基于同态加密算法的分布式加密流量分类隐私保护方法(Pa-Fed)。在分布式节点完成训练后,本地模型将参数通过Paillier同态加密传递至中心服务器端。在中心服务器进行参数聚合时,仍然维持参数的密文状态,以确保在传输过程中的隐私性。实验能够较好地保持分类精确率,并且在加密后对分布式节点数据进行梯度推理攻击,有效地验证了分布式节点数据的隐私性。

With the rapid development of information technology,the amount of data has increased rapidly,and the distributed storage methods have gradually evolved.To solve the problem that the distributed data storage mode is prone to gradient leakage caused by gradient inference attacks on model training,which in turn leads to the leakage of datasets in distributed nodes,a privacy protection method of distributed encrypted traffic classification(Pa-Fed)based on homomorphic encryption algorithm is proposed.After the distributed nodes are trained,the local model passes the parameters to the central server through Paillier homomorphic encryption.When the parameters are aggregated on the central server,the ciphertext state of the parameters is maintained to ensure privacy during transmission.The experiment can well maintain the classification accuracy rate,and carry out the gradient inference attack on the distributed node data after encryption,which effectively verifies the privacy of distributed node data.