Presentation 2012-11-07
Differential Privacy of Positive Semi-definite Matrices
Jun SAKUMA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We consider differential privacy of positive semidefinite matrices. We introduce two new randomization schemes; One is a mechanism which applies low-rank approximation as a post process after regular element-wise randomization with Laplace noises. The other is based on eigendecomposition using the matrix von-Mises Fisher distribution. Given a fixed privacy budget, we prove that our mechanisms achieve relatively better accuracy if lower-rank approximation of the output is acceptable. We experimentally demonstrate that low-rank approximation helps to control the accuracy-privacy trade-off with experiments of collaborative filtering and spectral network analysis.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) differential privacy / low-rank approximation / collaborative filtering / spectral analysis / graph Laplacian
Paper # IBISML2012-46
Date of Issue

Conference Information
Committee IBISML
Conference Date 2012/10/31(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Information-Based Induction Sciences and Machine Learning (IBISML)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Differential Privacy of Positive Semi-definite Matrices
Sub Title (in English)
Keyword(1) differential privacy
Keyword(2) low-rank approximation
Keyword(3) collaborative filtering
Keyword(4) spectral analysis
Keyword(5) graph Laplacian
1st Author's Name Jun SAKUMA
1st Author's Affiliation Graduate School oi SIE, University of Tsukuba:Japan Sciance and Technology Agency, PRESTO()
Date 2012-11-07
Paper # IBISML2012-46
Volume (vol) vol.112
Number (no) 279
Page pp.pp.-
#Pages 8
Date of Issue