| 本文已被：浏览 9348次 下载 18054次 |
码上扫一扫！ |
|
|
| 机器学习模型安全与隐私研究综述 |
|
纪守领1, 杜天宇1, 李进锋1, 沈超2, 李博3
|
|
1.浙江大学 网络空间安全研究中心 计算机科学与技术学院, 浙江 杭州 310027;2.智能网络与网络安全教育部重点实验室 电子与信息学部(西安交通大学), 陕西 西安 710049;3.Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana-Champaign, IL 61822, USA
|
|
| 摘要: |
| 在大数据时代下，深度学习、强化学习以及分布式学习等理论和技术取得的突破性进展，为机器学习提供了数据和算法层面强有力的支撑，同时促进了机器学习的规模化和产业化发展.然而，尽管机器学习模型在现实应用中有着出色的表现，但其本身仍然面临着诸多的安全威胁.机器学习在数据层、模型层以及应用层面临的安全和隐私威胁呈现出多样性、隐蔽性和动态演化的特点.机器学习的安全和隐私问题吸引了学术界和工业界的广泛关注，一大批学者分别从攻击和防御的角度对模型的安全和隐私问题进行了深入的研究，并且提出了一系列的攻防方法.回顾了机器学习的安全和隐私问题，并对现有的研究工作进行了系统的总结和科学的归纳，同时明确了当前研究的优势和不足.最后探讨了机器学习模型安全与隐私保护研究当前所面临的挑战以及未来潜在的研究方向，旨在为后续学者进一步推动机器学习模型安全与隐私保护研究的发展和应用提供指导. |
| 关键词: 机器学习 投毒攻击 对抗样例 模型隐私 人工智能安全 |
| DOI：10.13328/j.cnki.jos.006131 |
| 分类号: |
| 基金项目:国家重点研发计划（2018YFB0804102）；浙江省自然科学基金（LR19F020003）；浙江省科技计划（2019C01055）；国家自然科学基金（61772466，U1936215，U1836202，61822309，61773310，U1736205） |
|
| Security and Privacy of Machine Learning Models: A Survey |
|
JI Shou-Ling1, DU Tian-Yu1, LI Jin-Feng1, SHEN Chao2, LI Bo3
|
|
1.Institute of Cyberspace Research and College of Computer Science and Technology, Zhejiang University, Hangzhou 310027, China;2.Ministry of Education Key Laboratory for Intelligent Networks and Network Security and Faculty of Electronic and Information Engineering (Xi'an Jiaotong University), Xi'an 710049, China;3.Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana-Champaign, IL 61822, USA
|
| Abstract: |
| In the era of big data, breakthroughs in theories and technologies of deep learning, reinforcement learning, and distributed learning have provided strong support for machine learning at the data and the algorithm level, as well as have promoted the development of scale and industrialization of machine learning. However, though machine learning models have excellent performance in many real-world applications, they still suffer many security and privacy threats at the data, model, and application levels, which could be characterized by diversity, concealment, and dynamic evolution. The security and privacy issues of machine learning have attracted extensive attention from academia and industry. A large number of researchers have conducted in-depth research on the security and privacy issues of models from the perspective of attack and defense, and proposed a series of attack and defense methods. In this survey, the security and privacy issues of machine learning are reviewed, existing research work is systematically and scientifically summarized, and the advantages and disadvantages of current research are clarified. Finally, the current challenges and future research directions of machine learning model security and privacy research are explored, aiming to provide guidance for follow-up researchers to further promote the development and application of machine learning model security and privacy research. |
| Key words: machine learning poisoning attack adversarial example model privacy artificial intelligence security |
