New SF-GNN Framework Improves Fairness in Graph Neural Networks

Researchers have developed a new framework called Subgraph Filtering for Fair Graph Neural Networks (SF-GNN) to address unfairness in graph neural networks. This method tackles structural bias by identifying and filtering edges that propagate sensitive information, even when sensitive attributes are not explicitly used. SF-GNN selectively downweights or removes these bias-prone edges during the message-passing steps of GNNs. Experiments on multiple datasets demonstrate that SF-GNN improves fairness while maintaining strong predictive performance, offering a better balance between fairness and accuracy compared to existing methods. AI

IMPACT This research offers a novel approach to mitigate bias in graph neural networks, potentially improving their reliability and fairness in real-world applications.

RANK_REASON The cluster describes a new research paper detailing a novel framework for improving fairness in graph neural networks.

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• graph neural networks
• Hugging Face
• IArxiv
• ScienceCast
• SF-GNN
• edge filtering
• hub participation
• message passing
• sensitive homophily
• statistical-parity regularizer
• triadic closure

• paper
• safety

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →