PrivacyPreserving Approximation Center for Computational Mathematics Seminar UCSD
- Slides: 21
Privacy-Preserving Approximation Center for Computational Mathematics Seminar UCSD January 24, 2012 Olvi Mangasarian UW Madison & UCSD La Jolla Glenn Fung Siemens Medical Solutions, Malvern, PA
Problem Statement • Entities with related data points wish to obtain a function approximation based on all the data points • The entities are unwilling to reveal their data to each other • Each entity holds a different set of data points for all variables. Thus, the data is said to be horizontally partitioned • Our approach: privacy-preserving approximation (PPA) using random matrix transformations • PPA provides exact solution based on the total data • PPA does not reveal any private information
Outline • Horizontally partitioned approximation problem • Secure transformation via a random matrix • Method of solution: Privacy-preserving linear programming • Computational results • Summary
Horizontally Partitioned Example EPA wishes to compute total green house emission from three industries without revealing the data of each industry A 1 A 2 A 3
Horizontally Partitioned Data Matrix Variables 1 2. . …………. n 1 2. . Data Points. . . m A 1 A 2 A A 3
Secure Linear Approximation
Why Linear Approximation is Secure?
Linear Equation Representation
Linear Privacy Preserving Approximation Algorithm
Linear Privacy Preserving Approximation Algorithm (Continued)
Privacy Preserving Nonlinear Kernel Approximation
Nonlinear Privacy Preserving Approximation Algorithm
Nonlinear Privacy Preserving Approximation Algorithm (Continued)
Computatinal Results
Entities’ Regions
The Exact Sinc Function
Entity 1 Alone: Approximation to the sinc function
Entity 2 Alone: Approximation to the sinc function
Entities 1 & 2: Secure Approximation to the sinc function
Summary & Outlook Privacy preserving approximation of functions using privately held data: – Based on a transformation using a random matrix B – Get essentially exact approximation to the original function without revealing privately held data Possible extensions to problems with privately held constraints
References ftp: //ftp. cs. wisc. edu/pub/dmi/tech-reports/11 -04. pdf ftp: //ftp. cs. wisc. edu/pub/dmi/tech-reports/07 -03. pdf
- Using mathematics and computational thinking
- Computational mathematics
- Data center seminar
- Kentucky center of mathematics
- Kentucky center for mathematics
- Fspos
- Typiska drag för en novell
- Nationell inriktning för artificiell intelligens
- Vad står k.r.å.k.a.n för
- Varför kallas perioden 1918-1939 för mellankrigstiden
- En lathund för arbete med kontinuitetshantering
- Underlag för särskild löneskatt på pensionskostnader
- Tidböcker
- Sura för anatom
- Förklara densitet för barn
- Datorkunskap för nybörjare
- Boverket ka
- Tes debattartikel
- För och nackdelar med firo
- Nyckelkompetenser för livslångt lärande
- Påbyggnader för flakfordon
- Kraft per area