Disparity, Inequality, and Accuracy Tradeoffs in Graph Neural Networks for Node Classification

By Arpit Merchant et al

Published on Oct. 21, 2023

Read the original document by opening this link in a new tab.

Table of Contents

INTRODUCTION
PROBLEM SETUP
ALGORITHMS
PFR-AX
PostProcess
RELATED WORK
CONCLUSION
REFERENCES

Summary

The document discusses the tradeoffs in Graph Neural Networks (GNNs) for node classification, focusing on algorithmic fairness and accuracy. It introduces interventions like PFR-AX and PostProcess to mitigate biases in data and model predictions. The study evaluates these interventions on various datasets and GNN models, highlighting the need for balancing fairness and accuracy. Results show that no single intervention is universally optimal, with PFR-AX and PostProcess offering granular control over the tradeoff. The paper contributes to the field by addressing limitations in current approaches and proposing new interventions to improve model confidence in predicting positive outcomes for protected groups.