Kernel dependence regularizers and Gaussian processes with applications to algorithmic fairness
作者：
Highlights：
• A general framework of empirical risk minimization with fairness regularizers and an analysis of its risk and fairness statistical consistency results are presented.
• A Gaussian Process (GP) formulation of the fairness regularization framework is derived, which allows uncertainty quantification and principled hyperparameter selection.
• A normalized version of the fairness regularizer which makes it less sensitive to the choice of kernel parameters is derived.
摘要
•A general framework of empirical risk minimization with fairness regularizers and an analysis of its risk and fairness statistical consistency results are presented.•A Gaussian Process (GP) formulation of the fairness regularization framework is derived, which allows uncertainty quantification and principled hyperparameter selection.•A normalized version of the fairness regularizer which makes it less sensitive to the choice of kernel parameters is derived.
论文关键词：Fairness,Kernel methods,Gaussian processes,Regularization,Hilbert-Schmidt independence criterion
论文评审过程：Received 5 May 2020, Revised 7 March 2022, Accepted 21 July 2022, Available online 26 July 2022, Version of Record 14 August 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108922
