Title: Privacy friendly learning with empirical feature-based summary statistics
Authors: Huihui Qin - The Hong Kong Polytechnic University (China)
Xin Guo - The University of Queensland (Australia) [presenting]
Abstract: Nowadays the extensive collecting and analyzing of data is stimulating widespread privacy concerns, and therefore is increasing tensions between the potential sources of data and researchers. Obviously, a privacy-friendly learning framework can help to ease the tension, and to boost data-related research. We propose a new algorithm of regression learning with empirical features, which uses only summary statistics instead of raw data. The selection of empirical features serves as a trade-off between prediction precision and the protection of privacy. Mathematical analysis of the convergence of the framework is provided, which covers also the scenario where data sets are collected from different sources respectively.