Title: Differential Private Data Release for Mixed-type Data via Latent Factor Models

报告人： 张艳青

研究方向：应用统计、机器学习——差分隐私保护

Abstract:  The rise of big data processing, sharing and analysis makes the  protection of confidential information urgently necessary. Differential  privacy is a particular data privacy preserving technology which can  publish synthetic data or statistical analysis with a minimum disclo[1]sure  of private information of individual record. The tradeoff between  privacy-preserving and utility guarantee is always a challenge for  differential privacy technology, especially for synthetic data  generation. Moreover, mixed-type data containing continuous, ordinal  cate[1]gorical  and nominal data are becoming increasingly pervasive due to the rapid  development of various data collecting platforms. In this paper, we  propose a differential private data release algorithm for mixed-type  data with correlated dependency under the framework of latent factor  models. The proposed method can add a relatively small amount of noise  to synthetic data under the same level of privacy protection while  capturing correlation information. Moreover, the proposed algorithm can  generate synthetic data preserving the same data type as original data,  including categorical data, which greatly improves the utility of  synthetic data. The key idea of our method is to partially perturb the  projection of original data on perturbed eigenvector space to construct a  synthetic data generation model, and to utilize link functions between  discrete variables and continuous variables to ensure consistency of  synthetic data type with original data. The proposed method can generate  differentially private synthetic data at low computation cost even when  the origi[1]nal  data is high-dimensional. In theory, we establish differentially  private properties of the proposed method and upper bound on the utility  of synthetic data. Our numerical studies also demonstrate superb  performance of the proposed method on the utility guarantee of the  privacy-preserving data released.

报告题目：A Bridge Between Radar Signal Detection Problems and Mathematics and Statistics

报告人：荣尧

研究方向：统计学、数学与信息科学交叉学科

摘要：Radar  technology plays a critical role in numerous industries, such as  military and defense, navigation, aerospace and aviation, automotive,  remote sensing, and meteorology. Signal detection is one of the primary  functions of radar. This report aims to establish a strong link between  radar signal detection and mathematics and statistics by presenting our  research findings.