Privacy-Preserving Multi-label Classification Method

Multi-label learning is an important research field in machine learning,which has been paid wide attention in recent years.In multi-label learning,each training instance is associated with a set of labels to present its multiple semantic information,and the task is to predict the associated labels for each unclassified instance.Most noteworthy,the existing multi-label learning algorithms are only from the perspective of machine learning,they directly use the real feature information of training dataset when training classification model,and directly summit the unclassified instances to the classification model when classifying the unclassified instances.Then these schemes are only suitable for the owners of the training data set to train their own classification models,and then classify their own unclassified instances.However,the application scenario like this is very limited.If the owner of the training dataset and the owner of the unclassified instances are the two parties who do not trust each other,the existing multi-label learning schemes will cause the problem of privacy information leakage.Therefore,how to classify the instances at the same time to protect the data privacy information has become an urgent research direction.The main work of this paper is as follows:First,we study privacy-preserving multi-label classification in this paper.This paper combines the homomorphic encryption and the secure point product protocol under server-client model for multi-label classification,then we propose a method for privacy-preserving multi-label classification.The method makes both the client and the server fail to obtain any valuable privacy information about the other party.We prove the security of the method,analyze the computational and communication complexity of the method,and verify the efficiency of the method by simulation experiments.Second,for reducing the computing burden of clients in the classification process,we introduce two non-colluding cloud servers in the privacy-preserving multi-label classification method,and propose a privacy-preserving multi-label classification method over encrypted data in cloud.The proposed method uses the homomorphic encryption and a series of secure multiparty computing protocols to outsource the task of multi-label classification to the cloud servers.This method can not only protect the privacy information of clients and training dataset owners while completing multi-label classification tasks,but also greatly reduce the storage cost and calculation burden of them.We formally prove the security of our method,analyze the computational and communication complexity of the method,verify the efficiency of the method by simulation experiments.