Differential Privacy for Sparse Classification Learning
报告人:张海教授,西北大学 时间:2019年11月8号,9:30-10:30
Differential privacy is a notion of privacy that provides useful information while concealing the individual information. We present a differential privacy version of convex and nonconvexsparse classification approach. Based on alternating direction method of multiplier(ADMM) algorithm, we transform the solving of sparse problem into the multistepiteration process. Then we add exponential noise to stable steps to achieve privacyprotection. By the property of the post-processing holding of differential privacy, the proposed approach satisfies the differential privacy even when the originalproblem is unstable. Furthermore, we present the theoretical privacy bound of thedifferential privacy classification algorithm. Finally, we apply our framework to logistic regression with convex and nonconvex penalties.
张海,西北大学数学学院教授,统计系主任,博士生导师,现任陕西省统计学学会副理事长。博士毕业于西安交通大学,中国科学院数学与系统科学研究院博士后,美国加州大学伯克利分校统计系访问学者。目前科研主要研究涉及大数据学习算法、隐私保护和复杂系统及社会网络等领域。先后主持国家自然科学基金面上项目3项,主持国家自然科学基金广东大数据项目课题1项,参与国家重点基础研究发展规划(973计划)1项。在“Annals of Statistics”, “IEEE Transactions on Neural Networks and Learning System”, “Statistics and Probability Letters”,“中国科学(信息科学)”,“中国科学(数学)”,“数学学报”等杂志上发表科研论文30余篇