Research On Secure Distributed Machine Learning Based On Ring Signature

As AI has become a new focus of international competition and a new engine of economic development,AI strategy has become an important part of China’s path to building an innovative country and a world power in science and technology.Distributed machine learning has become a core means of strategic driving forces such as the mobile internet,big data,the Internet of Things,and supercomputing.At a time when distributed machine learning systems are favored by various industries,many potential security issues are gradually exposed.How to effectively protect personal privacy during the operation of a distributed machine learning system has become a challenge that every user needs to address in the era of big data.Based on this,this paper conducts the following research on privacy protection issues in distributed machine learning:First of all,this paper introduces the combined public key system in detail,points out and analyzes the security problems of the combined public key system,such as linear collusion attack,random public key replacement attack,which is no longer suitable for the generation of ring signature keys in distributed machine learning system.Based on the above analysis,this article designs a secure distributed machine learning combined public key generation algorithm,namely the SDCPK generation algorithm.In the key generation process,this paper solves the above attack problem by adding a random public private key pair generated by the user and the key management center,and binding the random public key with the user ID through the hash algorithm,and then applies it to the ring signature key generation and distributed machine learning transaction and training methods in this paper.Secondly,this paper designed a distributed machine learning data transaction method based on ring signature,and compared and analyzed its security,function and efficiency,aiming at the security issues of identity authentication,personal identity privacy protection,and the authentication process completely relying on a third-party certification authority,etc.Finally,this paper designs a distributed machine learning data training method based on ring signature,aiming at the problems of differential privacy protection of distributed machine learning data training,fixed learning leader,relying on traditional CA center and repeatedly sending training results to disrupt the generation of weight values,and verifies the feasibility of the method through simulation experiments.