Research On Data Computing Schemes For Privacy-Preserving In Internet Of Medical Things

The development of the Internet of Medical Things(Io MT)has not only accelerated the reform of the medical and healthcare system,but also improved the quality of medical services for residents and further promoted the rapid development of the medical industry.In the medical industry,sensor devices can transform real ”things” in ”physical space” into the”data” of ”virtual space”.By computing and analyzing data,a lot of deep value in the data can be uncovered.With sufficient data and abundant computing resources,models with better performance can be trained.However,medical data contains a large amount of private information,leading to the following three problems when using medical data for model training.(1)Due to the limited resources of some medical devices,some computing tasks need to be outsourced,but untrusted outsourced computing nodes pose a threat to privacy.(2)To ensure that the private information in the data is not leaked,data cannot be exchanged and shared between institutions,making it difficult to obtain comprehensive and sufficient data for training models.(3)Under the distributed condition,there is a certain correlation between the gradient and the training data in the model training process,and malicious users can recover the data by collecting the gradient information in the training process,resulting in data leakage.This dissertation addresses the aforementioned issues by combining the practical application requirements of privacy and data computation in the Io MT scenario,and investigates network security for three data application scenarios: ”centralised privacy-preserving data computation”,”privacy-preserving data computation under vertical federation”,and ”privacypreserving data computation under horizontal federation”.Meanwhile,an analysis of the technical solutions proposed by researchers,based on homomorphic encryption,secure multiparty computation,differential privacy,etc.,shows that these schemes still have problems,such as high computational resource consumption for local devices and affecting the accuracy of prediction results,which limits the deployment of some schemes.Unfortunately,the openness of the network also introduces many insecurities into the network,such as untrusted cloud service providers and malicious network nodes.The most important thing is that in the process of data processing,it is necessary to effectively detect the existence of insecure factors in the network,improve the level of data processing on the premise of ensuring the privacy and security of medical data,and provide better services to patients and other participants in the medical industry.Therefore,considering the limitations of current research and the requirements of privacy and data computation,the specific work of this dissertation is mainly divided into the following three aspects:(1)To solve the problem that there are dishonest nodes in the outsourcing computing scenario of Io MT,which leads to data leakage and computing task failure,a data security outsourcing computing scheme based on edge computing has been proposed.Based on the stochastic invertible matrix,noise is added to the user data and the user-side computational overhead is effectively reduced by offloading the high computational complexity user-side computational tasks to the edge computing node.In order to prevent dishonest edge nodes from influencing the computation results,a verifiable outsourcing computing algorithm is proposed that uses blocks to detect and record the malicious behavior of each participant.Experimental verification is carried out on the real data set to evaluate the safety and performance of the scheme,and the experimental results show that the proposed scheme is closer to the real application scenario in terms of safety and operational efficiency.(2)To solve the problems of data leakage and artificial intelligence model training failure caused by malicious nodes in vertical federation,a data privacy protection computing scheme for vertical federation is proposed.The scheme encrypts the data and sends it to the cloud for storage and computation,ensuring that the deployment and implementation of the scheme does not affect the normal business of each participant.Other participants in the scheme allow the data to be processed offline after it has been encrypted and stored,increasing the flexibility of the scheme.To evaluate the security and performance of the scheme,experimental verification is performed on real data sets,and the results show that the scheme significantly improves the flexibility of system deployment.(3)To solve the problem that the gradient of the exchange model in horizontal federation is susceptible to data leakage,a privacy-preserving computing scheme for horizontal federated learning has been proposed.To address the issue of gradient-induced data leakage in federated learning,this scheme proposes a method to generate dummy data based on GAN(Generative Adversarial Network)to construct a public dataset to effectively prevent privacy leakage.The teacher model uses the private data of participants and the public dataset for training,and guides the student model to use the public dataset for training through knowledge distillation technology.To effectively improve the prediction accuracy of the student model,based on blockchain technology,the teacher model uses miners to evaluate the student model,and eliminates inferior student models according to the model evaluation results to prevent them from negatively impacting the global model.This effectively reduces the impact of inferior models on the prediction accuracy of the global model and improves the accuracy of the global model.Experimental results demonstrate that the scheme can effectively resist attacks launched by adversaries in this scheme,and enhance the prediction accuracy of the global model.