Fairness Through Controlled (Un)Awareness in Node Embeddings

TL;DR

This paper explores how tuning the CrossWalk algorithm's parameters can control the detectability of sensitive attributes in node embeddings, aiding in creating fairer graph-based machine learning models.

Contribution

It introduces a method to adjust the fairness of node embeddings by hyperparameter tuning, balancing between attribute detectability and obfuscation.

Findings

Hyperparameter tuning affects attribute inference accuracy.

Controlled (un)awareness improves fairness in graph embeddings.

The method adapts to different fairness paradigms.

Abstract

Graph representation learning is central for the application of machine learning (ML) models to complex graphs, such as social networks. Ensuring `fair' representations is essential, due to the societal implications and the use of sensitive personal data. In this paper, we demonstrate how the parametrization of the \emph{CrossWalk} algorithm influences the ability to infer a sensitive attributes from node embeddings. By fine-tuning hyperparameters, we show that it is possible to either significantly enhance or obscure the detectability of these attributes. This functionality offers a valuable tool for improving the fairness of ML systems utilizing graph embeddings, making them adaptable to different fairness paradigms.