IMPROVING FAIRNESS THROUGH REINFORCING LEARNING

摘要

A method implemented on a computer for improving fairness in a supervised machine learning model can be provided. The method includes associating the supervised machine learning model with a metamodel for reinforcement learning, selecting a list of hyperparameters and parameters of the supervised machine learning model, and controlling at least one aspect of the supervised machine learning model by adjusting hyperparameter values and parameter values the list of hyperparameters and parameters of the supervised machine learning model by an reinforcement learning engine related to the reinforcement learning metamodel by calculating a reward function based on a plurality of conflicting objective functions. The method further comprises iteratively repeating the steps of selecting and controlling to improve a fairness value of the model for supervised machine learning.