2021/11/29 by Pavan Ravishankar, Ravishankar, Pavan, Pranshu Malviya +3 · 1 citation
Computer Science · Economics, Econometrics and Finance · Mathematics · Social Sciences · #Artificial intelligence #Computer science #Computers and Society (cs.CY) #Econometrics #Economics #Electoral Systems and Political Participation #Enhanced Data Rates for GSM Evolution #FOS: Computer and information sciences #Game Theory and Voting Systems #Graph #Legal and Constitutional Studies #Machine Learning (cs.LG) #Mathematical optimization #Mathematics #Parametric model #Parametric statistics #Prioritization #Statistics #Theoretical computer science #cs.CY #cs.LG
paper · pdf · doi:10.48550/arxiv.2111.14348
published in arXiv (Cornell University) (Cornell University) · ACML 2021
arxiv created 2021/11/29 · openalex publication_date 2021/11/29 · arxiv updated 2021/11/30 · openalex created_date 2022/07/25 · openalex updated_date 2026/08/05
In budget-constrained settings aimed at mitigating unfairness, like law\nenforcement, it is essential to prioritize the sources of unfairness before\ntaking measures to mitigate them in the real world. Unlike previous works,\nwhich only serve as a caution against possible discrimination and de-bias data\nafter data generation, this work provides a toolkit to mitigate unfairness\nduring data generation, given by the Unfair Edge Prioritization algorithm, in\naddition to de-biasing data after generation, given by the Discrimination\nRemoval algorithm. We assume that a non-parametric Markovian causal model\nrepresentative of the data generation procedure is given. The edges emanating\nfrom the sensitive nodes in the causal graph, such as race, are assumed to be\nthe sources of unfairness. We first quantify Edge Flow in any edge X -> Y,\nwhich is the belief of observing a specific value of Y due to the influence of\na specific value of X along X -> Y. We then quantify Edge Unfairness by\nformulating a non-parametric model in terms of edge flows. We then prove that\ncumulative unfairness towards sensitive groups in a decision, like race in a\nbail decision, is non-existent when edge unfairness is absent. We prove this\nresult for the non-trivial non-parametric model setting when the cumulative\nunfairness cannot be expressed in terms of edge unfairness. We then measure the\nPotential to mitigate the Cumulative Unfairness when edge unfairness is\ndecreased. Based on these measurements, we propose the Unfair Edge\nPrioritization algorithm that can then be used by policymakers. We also propose\nthe Discrimination Removal Procedure that de-biases a data distribution by\neliminating optimization constraints that grow exponentially in the number of\nsensitive attributes and values taken by them. Extensive experiments validate\nthe theorem and specifications used for quantifying the above measures.\n