Research And Implementation Of Copyright Protection For Multiuser Authorization Mechanism In Deep Neural Network Model

In recent years,deep neural network models have achieved great success in artificial intelligence applications such as image classification,target detection,speech recognition,and natural language processing.However,the deep neural network model is a highly complex computational model,and its training requires a large number of data sets,high computational cost and excellent algorithm ideas.Because the deep neural network model has high commercial value,its copyright protection is particularly important.In order to ensure sufficient protection of the copyright of deep neural network models,various technical means such as digital watermarking technology,encryption technology,access control technology,secure transmission technology,and secure storage technology need to be adopted.These technical means can provide more comprehensive and effective protection.In real scenes,we rarely have access to the internal machine of the deep neural network model.The reasoning function of the deep neural network model is mostly used by calling the application programming interface(API)of the deep neural network model,and the backdoor watermark is more suitable for commercial applications.However,there are still some problems in the application of backdoor watermarking.For example,the distribution difference between ordinary samples and key samples is large,so that the attacker can find the watermark and remove the watermark,thus undermining the copyright protection of the deep neural network model.In addition,after the attacker steals the deep neural network model,the owner of the deep neural network model still cannot forcibly terminate the continued use of the deep neural network model by the thief,which requires the government ’s law enforcement action to prevent the thief ’s infringement.Therefore,how to improve the robustness and reliability of backdoor watermarking,and how to strengthen the copyright protection of deep neural network model,remains an important issue facing current research.In view of the above problems,this paper studies and implements the copyright protection of the multiuser authorization mechanism of the deep neural network model.The main research contents involve the backdoor watermarking algorithm of the deep neural network model and the multiuser authorization framework of the deep neural network model.The backdoor watermarking algorithm of deep neural network model aims to solve the problem of large distribution difference between ordinary samples and key samples in the research of deep neural network model watermarking.The deep neural network model multiuser authorization framework aims to solve the problem that copyright infringement cannot be prevented in time.In order to verify the reliability of the multiuser authorization framework of the deep neural network model,we use MNIST,Fashion-MNIST,CIFAR-10,CIFAR-100 and Caltech-101 to verify the reliability of the multiuser authorization framework of the deep neural network model.Experiments were performed on data sets and more than a dozen deep neural network models.Experimental results show that our framework has higher effectiveness,concealment and fidelity,and can effectively authorize multiple users through authorization keys.In addition,our framework has no obvious side effects on the main functions of the original deep neural network model.