Privacy-preserving efficient subset selection of features for regression models in a multi-party computation setting
US11444926B1 · kind B1 · utility
Inventors
Key dates
| Filing date | Oct 15, 2019 |
| Grant date | Sep 13, 2022 |
| Priority date | — |
| Expiry date | Mar 1, 2041 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06N20/00
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
An efficient method of feature selection for regression models can be implemented in a privacy-preserving manner in a multi-party computation setting. In accordance with various embodiments, the method takes as input data a feature matrix, a dependent variable vector, and an external feature matrix from which a feature is to be selected for addition to a regression model. Some or all of the input data can include private data that can be secret shared during the method so as not to disclose the private data to other parties. Based on two heuristic assumptions, the method determines numerators and denominators for a t-statistics vector in multi-party computations and then calculates the t-statistics vector. In determining the numerators and denominators, the method can determine a baseline Hessian matrix and a vector of predictions. A feature represented in the external feature matrix is then selected based on the calculated t-statistics vector.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.