Authors: O. Deniz Kose, Yanning Shen

Published on: February 06, 2024

Impact Score: 8.2

Arxiv code: Arxiv:2402.04383

Summary

• What is new: This work introduces the first fair graph generation framework, FairWire, and a novel fairness regularizer to mitigate structural bias in both real and synthetic graphs.
• Why this is important: The existing machine learning models over graphs amplify disparate impact due to biased graph structures, affecting their real-world applicability.
• What the research proposes: A fairness regularizer and FairWire, a fair graph generation framework, have been proposed to reduce structural bias in graph-based machine learning.
• Results: Experimental validation on real-world networks showed significant mitigation of structural bias in both real and synthetic graphs.

Technical Details

Technological frameworks used: FairWire

Models used: Graph generation models with fairness regularizer

Data used: Real-world networks

Potential Impact

Companies relying on graph-based decision systems; markets in data privacy and machine learning as a service (MLaaS).

Want to implement this idea in a business?

We have generated a startup concept here: FairGraphAI.