B0606
Title: Differentially private projection depth-based medians
Authors: Kelly Ramsay - York University (Canada) [presenting]
Dylan Spicker - University of New Brunswick (Saint John) (Canada)
Abstract: Multivariate medians based on projection depth are popular robust location estimates. The propose-test-release framework offers a methodology for developing differentially private versions of robust statistics. The combination of these two techniques to produce approximately differentially private projection depth-based medians is explored. Both the probability of failing the test portion of the algorithm and the accuracy-privacy trade-off are quantified under general distributional assumptions. Examples of applying such theory to specific projection depth-based medians are discussed. The findings highlight the connection between the probability of passing the 'test' in the propose-test-release approach and the estimate's gross error sensitivity.
