Research On Security Technology Of Federated Learning For Smart Healthcare

With the vigorous development of the Internet of Things(Io T)and artificial intelligence technology,Federated Learning(FL)technology provides a new solution for protecting the privacy of smart healthcare data.However,the existing FL technology still has security and efficiency issues.On the one hand,in FL,although raw data does not need to be transmitted to the central server,sensitive information can still be calculated from uploaded local model parameters.Therefore,secure model aggregation algorithms are needed to ensure the security of model parameters.On the other hand,the communication cost and computation overhead of existing solutions need to be reduced due to frequent secret sharing.In addition,to ensure that the data owned by the participants in FL meet higher quality requirements based on compliance,the authenticity and credibility of local data need to be verified.It can be seen that FL provides a new solution for protecting the privacy of intelligent medical data,but further research and improvement are needed.To address the above issues,this thesis conducted the following research:1.To verify that the user device that uploaded data is deployed by the system itself under the premise of protecting user privacy data,and reducing the time cost of system operation,this thesis proposes a federated learning scheme based on full dynamic secret sharing.In this scheme,the double-mask protocol is used to keep the user’s local model parameters secret,and the homogeneous linear recursive equation and elliptic curve cryptosystem are used to realize fully dynamic secret sharing and user identity authentication.Additionally,the scheme allows users to join or exit FL during the process.Finally,the security and performance of the scheme were analyzed,and the results showed that the scheme is secure and improves efficiency by 30% and 60% respectively under the two scenarios of users dropping out and not dropping out.Overall,the new scheme provides a feasible solution to protect data privacy and ensure data processing efficiency when collecting user health data in the Io T smart healthcare system.2.To select FL participants securely and efficiently in a public environment,while motivating users to voluntarily upload high-quality and compliant model parameters,this thesis proposes a federated learning scheme based on fine-grained collaborative access control.The scheme is based on a fine-grained collaborative access control mechanism and uses Attribute-Based Encryption(ABE)to implement the selection of specific data types or medical institutions participating in FL.In this scheme,the data attributes of medical institutions can be used as access control conditions,and only institutions with corresponding attributes can participate in FL.In addition,the scheme uses a blockchainbased incentive mechanism to enhance the willingness of medical institutions to share highquality model parameters and participate in FL algorithms.Using blockchain technology,medical institutions can record the quality of the contributed model parameters in FL algorithms and obtain corresponding incentives.Finally,the scheme uses smart contracts to ensure the fairness of the incentive mechanism.The results of security and performance analysis show that the scheme slightly improves efficiency while ensuring security.In summary,the new scheme achieves a secure and efficient FL scheme through fine-grained collaborative access control,blockchain-based incentive mechanism,and smart contract technology,which provides a new solution for data sharing and collaboration in the field of smart healthcare.