{ "cells": [ { "cell_type": "code", "execution_count": 7, "metadata": { "collapsed": true }, "outputs": [], "source": [ "def test(p1,p2,epsilon):\n", " for testVal in np.arange(0.5,10,0.5):\n", " a = p2.pdf(testVal)\n", " b1 = p1.pdf(testVal)\n", " b2 = np.exp(epsilon)\n", " b = b1*b2" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#Differential Privacy\n", "\n", "##Motivation\n", "\n", "Quoting Paul Ohm:\n", "\n", ">In the 1990s, an insurance company released 'anonymised' health data on **all state employees** in Massachusetts. The Governor of Massachusetts, said that GIC had protected patient privacy by deleting identifiers. Latanya Sweeney (a postgrad researcher) showed otherwise.\n", "\n", "How it was done:\n", " - She knew that Governor Weld resided in Cambridge, Massachusetts.\n", " - Used the voter rolls from the city of Cambridge (contains the name, address, ZIP code, birth date, and sex of every voter).\n", " - By combining this data with the GIC records, Sweeney found Governor Weld with ease.\n", " - Only six people in Cambridge shared his birth date (contained in the 'anonymised' data)\n", " - Only three of them men (also in the data)\n", " - Of them, only he lived in his ZIP code (also in the data)\n", "\n", "(An example of a linkage attack)\n", "\n", "