US 12,287,848 B2
Learning Mahalanobis distance metrics from data
Mikhail Yurochkin, Cambridge, MA (US); Debarghya Mukherjee, Kolkata (IN); Moulinath Banerjee, Portage, MI (US); Yuekai Sun, Ann Arbor, MI (US); and Sohini Upadhyay, Cambridge, MA (US)
Assigned to International Business Machines Corporation, Armonk, NY (US); and Regents of the University of Michigan, Ann Arbor, MI (US)
Filed by International Business Machines Corporation, Armonk, NY (US); and REGENTS OF THE UNIVERSITY OF MICHIGAN, Ann Arbor, MI (US)
Filed on Jun. 11, 2021, as Appl. No. 17/345,730.
Prior Publication US 2022/0405529 A1, Dec. 22, 2022
Int. Cl. G06F 18/22 (2023.01); G06F 18/20 (2023.01); G06F 18/214 (2023.01); G06F 18/40 (2023.01); G06N 3/08 (2023.01)
CPC G06F 18/22 (2023.01) [G06F 18/214 (2023.01); G06F 18/285 (2023.01); G06F 18/40 (2023.01); G06N 3/08 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for improving algorithmic fairness of machine learning models using learned fair Mahalanobis distance similarity metrics, the method comprising:
obtaining training data comprising similarity annotations;
determining one model out of a plurality of models to use in learning a Mahalanobis covariance matrix Σ based on the obtained training data;
learning the Mahalanobis covariance matrix Σ from the obtained training data using the determined one model, wherein the Mahalanobis covariance matrix Σ represents a fair Mahalanobis distance similarity metric; and
training one or more machine learning models using, at least in part, the fair Mahalanobis distance similarity metric, for one or more machine learning model tasks.