Optimal Transport of Classifiers to Fairness

Maarten, Buyl,; Bie, Tijl De

doi:10.48550/arxiv.2202.03814

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the second approach, one can use the theory of optimal transport to find the distribution of scores corresponding to all the sensitive populations and optimally transport them to a fair score distribution. This approach has been followed in papers such as [36][37][38][39][40] . In this paper, we apply the score transformation approach to graph data by taking the former route and using quadratic programming to transform the scores to fair ones.…”

Section: Score Transformationmentioning

confidence: 99%

Fair Link Prediction with Overlapping Groups

Pal,

Sikdar,

Ganguly

2023

Preprint

View full text Add to dashboard Cite

In this paper, we consider the problem of ensuring fairness in graph data, more specifically the implementation of fair link prediction. In the context of link prediction in graphs, the fairness notions of demographic parity and equalized odds tries to ensure equal average linking probability and true positive rates across different demographic groups consisting of various node pairs. In this paper, we propose novel algorithms to implement these fairness notions while taking overlapping and intersectional groups into account that arise in the case of multiple sensitive attributes, unlike most of the existing methods which only consider a single sensitive attribute. Assuming that the link prediction model generates a prediction score for each node pair to form an edge, we formulate a convex optimization problem that minimizes the squared Euclidean distance between the original prediction scores and transformed scores, subject to the fairness constraints. The transformed scores are then utilized for fair link prediction. We refer to our overall framework as Fair Link Prediction for Graphs (FairLPG) and demonstrate our method's effectiveness on real-world datasets and graph neural network based link predictions models. It outperforms or performs competitively with existing methods both in terms of fairness and prediction accuracy. To the best of our knowledge, this work is the first to handle the case of intersectional sensitive groups in the graph setting.

show abstract