Enhanced Synthetic Data and a Unified Framework for Quantifying Privacy Risk in Synthetic Data
Embodiments disclosed herein improve data privacy and security by combining synthetic data and statutory pseudonymization to create protected data that is more effectively disconnected from the original source data. By bringing synthetic data and statutory pseudonymization techniques together, a flexible level of protection may be applied to data that strikes an appropriate balance between the ease of use of cleartext data and the aggressive protection of statutory pseudonymization. Further embodiments disclosed herein improve data privacy and security by providing a novel statistical framework that jointly quantifies different types of privacy risks in synthetic datasets and that includes attack-based evaluations for the singling out, linkability, and inference risks. According to other embodiments, the modular nature of the framework facilitates the future integration of new and potentially stronger attacks for evaluating privacy risks. The framework separates the evaluation of the success rate of the privacy attacks from the calculation of the reported privacy risks.
1 . A system, comprising:
a memory having, stored therein, computer program code; and
one or more processing units operatively coupled to the memory and configured to execute instructions in the computer program code that cause the one or more processing units to:
obtain a first cleartext dataset comprising a first plurality of datum;
generate a first synthetic dataset based, at least in part, on the first cleartext dataset, wherein the first synthetic dataset comprises a second plurality of datum;
apply at least one pseudonymization technique to at least one datum in the first synthetic dataset to generate a first enhanced synthetic dataset; and
perform at least one analysis operation on the first enhanced synthetic dataset.
2 . The system of claim 1 , wherein the instructions in the computer program code further cause the one or more processing units to:
transmit the first enhanced synthetic dataset to a third-party.
3 . The system of claim 1 , wherein the instructions that cause the one or more processing units to apply at least one pseudonymization technique to at least one datum in the first synthetic dataset further comprise instructions that cause the one or more processing units to:
pseudonymize at least one field name in the first synthetic dataset.
4 . The system of claim 1 , wherein the instructions that cause the one or more processing units to apply at least one pseudonymization technique to at least one datum in the first synthetic dataset further comprise instructions that cause the one or more processing units to:
perform a generalization operation on at least one field in the first synthetic dataset.
5 . The system of claim 1 , wherein the instructions that cause the one or more processing units to apply at least one pseudonymization technique to at least one datum in the first synthetic dataset further comprise instructions that cause the one or more processing units to:
apply at least one pseudonymization technique to fewer than all of the second plurality of datum in the first synthetic dataset.
6 . The system of claim 1 , wherein the instructions in the computer program code further cause the one or more processing units to:
train a machine learning (ML) model with the first enhanced synthetic dataset.
7 . The system of claim 6 , wherein the instructions in the computer program code further cause the one or more processing units to:
use the trained ML model to restore the first enhanced synthetic dataset to a cleartext dataset.
8 . A system, comprising:
a memory having, stored therein, computer program code; and
one or more processing units operatively coupled to the memory and configured to execute instructions in the computer program code that cause the one or more processing units to:
obtain a first synthetic dataset comprising a first plurality of datum;
perform one or more privacy attacks on the first synthetic data;
measure a success rate for each of the one or more privacy attacks;
quantify a risk level for each of the one or more privacy attacks based, at least in part, on the respective success rate for the privacy attack; and
output the quantified risk level for at least one of the one or more privacy attacks.
9 . The system of claim 8 , wherein the instructions in the computer program code comprise part of a modular privacy framework.
10 . The system of claim 8 , wherein the instructions that cause the one or more processing units to quantify a risk level for each of the one or more privacy attacks further comprise instructions that cause the one or more processing units to:
compare results of the respective privacy attack against at least two baselines.
11 . The system of claim 10 , wherein a first baseline of the at least two baselines comprises performing the respective privacy attack on a control dataset from a same distribution as the first synthetic dataset.
12 . The system of claim 11 , wherein a second baseline of the at least two baselines comprises performing a randomly-determined privacy attack on the first synthetic dataset.
13 . The system of claim 8 , wherein the one or more privacy attacks comprise one or more of:
a singling out attack; a linkability attack; or an inference attack.
14 . A computer-implemented method, comprising:
obtaining, a first framework, a first cleartext dataset comprising a first plurality of datum;
generating, by the framework, a first synthetic dataset based, at least in part, on the first cleartext dataset, wherein the first synthetic dataset comprises a second plurality of datum;
applying, by the framework, at least one pseudonymization technique to at least one datum in the first synthetic dataset to generate a first enhanced synthetic dataset; and
performing, by the framework, at least one analysis operation on the first enhanced synthetic dataset.
15 . The computer-implemented method of claim 14 , further comprising:
transmitting, by the framework, the first enhanced synthetic dataset to a third-party.
16 . The computer-implemented method of claim 14 , wherein applying, by the framework, at least one pseudonymization technique to at least one datum in the first synthetic dataset further comprises:
pseudonymizing at least one field name in the first synthetic dataset.
17 . The computer-implemented method of claim 14 , wherein applying, by the framework, at least one pseudonymization technique to at least one datum in the first synthetic dataset further comprises:
performing a generalization operation on at least one field in the first synthetic dataset.
18 . The computer-implemented method of claim 14 , wherein applying, by the framework, at least one pseudonymization technique to at least one datum in the first synthetic dataset further comprises:
applying at least one pseudonymization technique to fewer than all of the second plurality of datum in the first synthetic dataset.
19 . The computer-implemented method of claim 14 , further comprising:
training a machine learning (ML) model with the first enhanced synthetic dataset.
20 . The computer-implemented method of claim 19 , further comprising:
using the trained ML model to restore the first enhanced synthetic dataset to a cleartext dataset.