Research On Federated Learning Based On Integrated Similarity Node Blockchain

Currently,artificial intelligence(AI)technology is rapidly developing worldwide with broad applications in fields such as healthcare,finance,transportation,education,and more.In the process of applying AI,a large amount of personal data must be collected,including identification information,health data,social network data,and more.This data may be abused or leaked,causing harm to users’ privacy.At present,challenges and shortcomings still exist in the area of data privacy protection for AI technology,which requires further strengthening of relevant laws,regulations,and regulatory mechanisms to better protect users’ data privacy.Federated learning can effectively avoid external infringement of user information,allowing for better cultivation and development of deep learning algorithms and overcoming the challenge of information isolation.Compared to traditional centralized models,federated learning is more flexible and able to cope with various complex environments.However,in the case of malicious node attacks,the security and robustness of traditional federated learning systems are challenged and may face enormous pressure from network transmission,as well as possible impact from malicious data.The research in this paper includes the following:(1)In order to solve the problem of node voting passivity and malicious manipulation of election results in PoS consensus mechanism,a co-neighboring node similarity model is introduced to achieve community partitioning,shorten the voting period,and improve consensus efficiency.Secondly,the credibility of each node is calculated,and the highest credibility node is selected as the witnessing node responsible for block production for each community.The node category is updated in a timely manner through the node identity conversion mechanism.(2)This paper improves the SC-DPoS node trust evaluation model based on the "reward mechanism" and similarity node screening to solve the problems of node activity and bribery risk in the PoS consensus process.Experimental results show that applying game theory incentive mechanism in the SC-DPoS consensus mechanism can effectively address the issue of inactive voting,allowing nodes to participate effectively in voting and ensure additional profits,thus enhancing the security and effectiveness of wireless network transmission.(3)A federated learning algorithm based on the SC＿DPoS consensus mechanism is designed.The reputation system is combined with the reward mechanism to dynamically adjust the allocation of node reward sharing,achieving a more specific and fair distribution of rewards to nodes.This mechanism solves the problems of single point of failure,centralization,and privacy leakage in updating global models during federated learning.Under this mechanism,the quantity and quality of data contributed by each node are improved,and the system stability is also better.This paper aims to improve the security of federated learning data transmission by introducing an incentive mechanism based on blockchain technology and implementing a federated learning framework supported by the reputation mechanism of the blockchain platform.Additionally,a community was formed through similarity calculation selection,achieving fairness in reward distribution and improving the stability and quality of the system.