Methods, systems, and computer readable media for applying pairwise differential privacy to variables in a data set

A system for applying pairwise differential privacy to variables in a data set, the system comprising:

• a computing platform including at least one processor and a memory;
• and
• a pairwise differential privacy (PDP) engine that includes a correlation manager and a noise generation manager (NGM) and is stored in the memory and when executed by the at least one processor is configured to:
• receive as input, a dataset comprising a plurality of variables having values, wherein the dataset comprises human physical or biological measurement data generated by a laboratory;
• store the values in a data table;
• store, in a correlation preservation table, identities of variables in the dataset that are correlated;
• execute, by the correlation manager, an algorithm that determines, for each variable represented by values in the data table, including a first variable represented by the values in the data table, whether the first variable is identified in the correlation preservation table, and, in response to determining that the first variable is identified in the correlation preservation table, reads, from the correlation preservation table, identities of variables that are correlated with the first variable and assigns a random instance seed value to the first variable and the variables identified as being associated with the first variable;
• compute, utilizing the correlation manager, values of an adaptive sensitivity parameter corresponding to values of the first variable and the variables identified as being correlated with the first variable, wherein computing the values of the adaptive sensitivity parameter includes inputting the values of the first variable and the variables identified as being correlated with the first variable into a relative sensitivity function that uses a conditional structure to compute and output the values of the adaptive sensitivity parameter, wherein the values of the adaptive sensitivity parameter vary, according to the conditional structure, with the values of the first variable and the variables identified as being correlated with the first variable and thereby providing pairwise differential privacy to the values of the first variable and the variables identified as being correlated with the first variable;
• utilize, by the noise generation manager, the random instance seed value and the values of the adaptive sensitivity parameter to determine, generate, and apply different amounts of additive noise to the values of the first variable and the variables identified as being correlated with the first variable to produce pseudonymized values of the first variable and the variables identified as being correlated with the first variable, wherein the noise generation manager includes a Laplace function that receives, as inputs, the values of the first variable and the variables identified as being correlated with the first variable, the values of the adaptive sensitivity parameter, and the random instance seed value and produces, as outputs the pseudonymized values of the first variable and the variables identified as being correlated with the first variable;
• and
• communicate the pseudonymized values of the first variable and the variables identified as being correlated with the first variable from the laboratory to a computer associated with a statistics team for identifying statistical trends represented by the pseudonymized values, wherein communicating the pseudonymized values of the first variable and the variables identified as being correlated with the first variable includes communicating the pseudonymized values of the first variable and the variables identified as being correlated with the first variable from a client-side host to a data management host over a network;
• and
• a pseudorandom data manager implemented on the data management host for generating a statistical report using the pseudonymized values of the first variable and the variables identified as being correlated with the first variable, wherein the statistical report indicates statistical relationships among the pseudonymized values of the first variable and the variables identified as being correlated with the first variable that are consistent with statistical relationships among the values of the first variable and the variables identified as being correlated with the first variable in the dataset.

A non-transitory computer readable medium having stored thereon executable instructions that when executed by a processor of a computer control the computer to perform a method comprising:

• receiving as input, a dataset comprising a plurality of variables having values, wherein the dataset comprises human physical or biological measurement data generated by a laboratory;
• storing the values in a data table;
• storing, in a correlation preservation table, identities of the variables that are correlated;
• executing an algorithm that determines, for each variable represented by values in the data table, including a first variable represented by the values in the data table, whether the first variable is identified in the correlation preservation table, and, in response to determining that the first variable is identified in the correlation preservation table, reads, from the correlation preservation table, identities of variables that are correlated with the first variable;
• computing values of an adaptive sensitivity parameter for values of the first variable and the variables identified as being correlated with the first variable, wherein computing the values of the adaptive sensitivity parameter includes inputting the values of the first variable and the variables identified as being correlated with the first variable into a relative sensitivity function that uses a conditional structure to compute and output the values of the adaptive sensitivity parameter, wherein the values of the adaptive sensitivity parameter vary, according to the conditional structure, with the values of the first variable and the variables identified as being correlated with the first variable;
• utilizing, by a noise generation manager, the random instance seed value and the values of the adaptive sensitivity parameter to determine, generate, and apply different amounts of additive noise to the values of the first variable and the variables identified as being correlated with the first variable to produce pseudonymized values of the first variable and the variables identified as being correlated with the first variable and which preserve correlations between the first variable and the variables identified as being correlated with the first variable, wherein the noise generation function includes a Laplace function that receives, as inputs, the values of the first variable and the variables identified as being correlated with the first variable, the values of the adaptive sensitivity parameter, and the random instance seed value and produces, as outputs the pseudonymized values of the first variable and the variables identified as being correlated with the first variable and thereby providing pairwise differential privacy to the values of the first variable and the variables identified as being correlated with the first variable;
• communicating the pseudonymized values of the first variable and the variables identified as being correlated with the first variable from the laboratory to a computer associated with a statistics team for identifying statistical trends represented by the pseudonymized values, wherein communicating the pseudonymized values of the first variable and the variables identified as being correlated with the first variable includes communicating the pseudonymized values of the first variable and the variables identified as being correlated with the first variable from a client-side host to a data management host over a network;
• and
• generating, by a pseudorandom data manager implemented on the data management host, a statistical report using the pseudonymized values of the first variable and the variables identified as being correlated with the first variable, wherein the statistical report indicates statistical relationships among the pseudonymized values of the first variable and the variables identified as being correlated with the first variable that are consistent with statistical relationships among the values of the first variable and the variables identified as being correlated with the first variable in the dataset.