面向用户数据和模型数据的隐私保护技术  

Privacy-Preserving Technology for User Data and Model Data

作　　者：陈嘉乐 张佳乐 杨子路 赵彦超[1] 后弘毅 陈兵[1] CHEN Jiale;ZHANG Jiale;YANG Zilu;ZHAO Yanchao;HOU Hongyi;CHEN Bing(School of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210023,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京211106 [2]中国电子科技集团公司第二十八研究所,南京210023

出　　处：《指挥信息系统与技术》2022年第6期95-100,共6页Command Information System and Technology

基　　金：国家重点研发计划(2017YFB0802300);中央高校基本科研业务费专项和南京航空航天大学研究生创新基地(实验室)开放基金(Kfjj20191607)资助项目。

摘　　要：数据与模型作为人工智能框架的2个重要元素,既需要考虑它们的安全性与隐私需求,又需要兼顾它们的计算性能。将密码学与人工智能技术相结合,面向多种智能算法,通过安全多方计算技术重构神经网络核心算法,实现全流程的安全推理。基于多线程计算和分段加密等技术优化Blowfish算法,高效完成海量模型数据的加解密流程。该研究有助于提升人工智能模式的安全性与密文计算效率,具有重要研究意义。The data and the model are the two most important elements of an artificial intelligence framework, their security and privacy requirements are the primary consideration, they also need to maintain the computing performance. Based on the combination of cryptography and artificial intelligence technology, the core algorithms of neural network is reconstructed by the secure multi-party computing technology to realize the secure reasoning the whole process. The Blowfish algorithms is optimized based on the multi-threaded computation and the segmented encryption technology, thus it can efficiently complete the encryption and decryption process of massive model data. This research is helpful to improve the security of artificial intelligence mode and the efficiency of cryptographic computation, and it has important research significance.

关 键 词：隐私保护深度学习 安全多方计算 BLOWFISH算法 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]