FairGC: Fairness-aware Graph Condensation

Dimensions overall score 7.0

• FairGC significantly reduces disparity in Statistical Parity (ΔSP) and Equal Opportunity (ΔEO) compared to existing state-of-the-art graph condensation models.result 90%
• Current graph condensation methods are bias-blind and frequently capture and amplify demographic disparities found in the original data.limitation 85%
• FairGC embeds fairness directly into the graph distillation process via a unified framework with three key components: a Distribution-Preserving Condensation module, a Spectral Encoding module, and a Fairness-Enhanced Neural Architecture.method 90%
• The Distribution-Preserving Condensation module synchronizes the joint distributions of labels and sensitive attributes to prevent bias propagation.method 90%
• The Spectral Encoding module uses Laplacian eigen-decomposition to preserve essential global structural patterns.technical 90%
• Existing fairness-aware graph models (e.g., FMP) lack robustness for structural reduction, treating fairness and data reduction as isolated tasks.limitation 80%
• FairGC provides a superior balance between accuracy and fairness compared to existing methods, as shown on four real-world datasets.result 85%
• Graph condensation (GC) distills large-scale graph structures into a significantly smaller set of synthetic nodes, ensuring models trained on the compact proxy achieve comparable performance to those trained on the full dataset.technical 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card