Iterative Orthogonal Feature Projection for Diagnosing Bias in Black-Box Models

2016/11/15 by Julius Adebayo, Lalana Kagal, Adebayo, Julius +1 · 7 citations
Computer Science · Economics, Econometrics and Finance · Mathematics · Social Sciences · #Ethics and Social Impacts of AI #Imbalanced Data Classification Techniques #Law, Economics, and Judicial Systems #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.1611.04967

arxiv created 2016/11/15 · arxiv updated 2016/11/16

Abstract

Predictive models are increasingly deployed for the purpose of determining access to services such as credit, insurance, and employment. Despite potential gains in productivity and efficiency, several potential problems have yet to be addressed, particularly the potential for unintentional discrimination. We present an iterative procedure, based on orthogonal projection of input attributes, for enabling interpretability of black-box predictive models. Through our iterative procedure, one can quantify the relative dependence of a black-box model on its input attributes.The relative significance of the inputs to a predictive model can then be used to assess the fairness (or discriminatory extent) of such a model.

Cited by

• Algorithm Fairness in AI for Medicine and Healthcare
• Fairkit, Fairkit, on the Wall, Who's the Fairest of Them All? Supporting Data Scientists in Training Fair Models
• Automated Test Generation to Detect Individual Discrimination in AI Models
• The Authority of "Fair" in Machine Learning
• Empirical Risk Minimization under Fairness Constraints
• Kernel Dependence Regularizers and Gaussian Processes with Applications to Algorithmic Fairness
• Discrimination-free Insurance Pricing with Privatized Sensitive Attributes
• Distill-and-Compare

Related

• Thinking Outside the Box: Orthogonal Approach to Equalizing Protected Attributes
• Auditing Black-box Models for Indirect Influence
• Simulating Biases for Interpretable Fairness in Offline and Online Classifiers
• Interpretable Companions for Black-Box Models
• Why Is My Classifier Discriminatory?