US 12,282,836 B2
Learning robust predictors using game theory
Kartik Ahuja, White Plains, NY (US); Karthikeyan Shanmugam, Elmsford, NY (US); Kush Raj Varshney, Ossining, NY (US); and Amit Dhurandhar, Yorktown Heights, NY (US)
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on Dec. 8, 2020, as Appl. No. 17/115,489.
Prior Publication US 2022/0180254 A1, Jun. 9, 2022
Int. Cl. G06N 20/20 (2019.01); G06N 7/00 (2023.01)
CPC G06N 20/20 (2019.01) [G06N 7/00 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
defining, by a computing device, a plurality of environment-specific classifiers corresponding to a plurality of environments;
constructing, by the computing device, an ensemble classifier associated with the plurality of environment-specific classifiers;
initiating, by the computing device, a game including a plurality of players corresponding to the plurality of environments, wherein the initiated game comprises;
training, by the computing device, the constructed ensemble classifier by each environment of the plurality of environments;
modifying, by the computing device, a respective environment-specific classifier by a corresponding environment of the plurality of environments based on feedback to improve performance of the modified respective environment-specific classifier in the corresponding environment; and
updating, by the computing device, the trained constructed ensemble classifier based on the modified respective environment-specific classifier;
calculating, by the computing device, a nash equilibrium of the initiated game;
determining, by the computing device, an ensemble predictor based on the calculated nash equilibrium of the initiated game; and
deploying, by the computing device, the determined ensemble predictor as an invariant predictor to make predictions in a new environment.