Privacy-preserving efficient subset selection of features for regression models in a multi-party computation setting

A computer system comprising at least one processor and a memory storing instructions that cause the computer system to perform, in coordination with other computer systems operating in a secure multi-party privacy-preserving computation setting, a method comprising:

• accessing a set of input data comprising a baseline feature matrix, a dependent variable vector, and an external feature matrix, wherein a portion of the input data is represented as secret shared data;
• performing a regression on the baseline feature matrix and the dependent variable vector to determine a coefficients vector baseline model;
• determining numerators for a t-statistics vector based on the baseline feature matrix, the dependent variable vector, the baseline model, and the external feature matrix in a multi-party computation;
• determining denominators for the t-statistics vector based on the baseline feature matrix, the baseline model, and the external feature matrix in a multi-party computation;
• computing the t-statistics vector based on the numerators and the denominators for the t-statistics vector;
• and
• selecting an external feature represented in the external feature matrix for addition to the baseline feature matrix based on the t-statistics vector,
• wherein a number of rounds of communication required between parties in the multi-party computations is independent of a size of the external feature matrix.

A non-transitory computer readable medium having instructions stored thereon, wherein the instructions, when executed by a computer system having at least one processor, cause the computer system to perform, in coordination with other computer systems operating in a secure multi-party privacy-preserving computation setting, a method comprising:

• accessing a set of input data comprising a baseline feature matrix, a dependent variable vector, and an external feature matrix, wherein a portion of the input data is represented as secret shared data;
• performing a regression on the baseline feature matrix and the dependent variable vector to determine a coefficients vector baseline model;
• determining numerators for a t-statistics vector based on the baseline feature matrix, the dependent variable vector, the baseline model, and the external feature matrix in a multi-party computation;
• determining denominators for the t-statistics vector based on the baseline feature matrix, the baseline model, and the external feature matrix in a multi-party computation;
• computing the t-statistics vector based on the numerators and the denominators for the t-statistics vector;
• and
• selecting an external feature represented in the external feature matrix for addition to the baseline feature matrix based on the t-statistics vector,
• wherein a number of rounds of communication required between parties in the multi-party computations is independent of a size of the external feature matrix.