Author: Zen and the Art of Computer Programming
On the basis of traditional information security, the field of artificial intelligence has paid more and more attention to privacy protection in recent years. However, due to the widespread defects of the current artificial intelligence models, serious problems such as privacy leakage will also occur in practical applications. Therefore, there are higher requirements for the security and privacy protection of artificial intelligence systems and services.
With the continuous development of artificial intelligence technology, how to ensure the security and privacy protection of artificial intelligence systems has become an important issue. At present, mainstream artificial intelligence privacy protection related research mainly focuses on the following three aspects:
- Data security protection. Protect the true and legitimate owners of training data, development data, and models. Prevent data from being stolen or tampered with, and ensure data security, confidentiality, and privacy.
- Model security protection. Based on the theories of trusted computing and safe model operation, technical means based on security boundaries and model attack detection are proposed to protect the model.
- Service Security Protection. When deploying artificial intelligence services to the actual production environment, it is necessary to consider the security of the service, including ensuring the availability, disaster recovery capability, and robustness of the service. At the same time, it is also necessary to pay attention to the privacy protection of data, especially when sensitive data is involved in artificial intelligence services.
However, no matter what kind of program, they are facing different degrees of problems. For example, after the model is deployed, personal data leakage and hacking attacks that are sensitive to users may still occur. In addition, these solutions often focus on a certain type of specific technical means, and do not solve the privacy protection problem as a whole. Therefore, in order to better ensure the security and privacy protection of artificial intelligence systems, it is of great significance to formulate corresponding technical specifications or standards.
This article will introduce the latest research results in the field of artificial intelligence privacy protection from the perspective of machine learning/deep learning framework. Firstly, the common concepts and terms in artificial intelligence privacy protection are introduced, and then the technical solutions of data security protection, model security protection and service security protection are respectively explained, and the relevant future development trends and challenges are given at last.
2. Explanation of basic concepts and terms
2.1 Data Security Protection
Data security refers to protecting the true and legal owners of training data, development data and models&