HeteroFedSyn: Differentially Private Tabular Data Synthesis for Heterogeneous Federated Settings

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• HeteroFedSyn is the first differentially private tabular data synthesis framework designed specifically for horizontal federated settings.method 95%
• Existing DP tabular synthesis methods predominantly assume centralized or local settings and overlook horizontal federated scenarios.limitation 90%
• Naively synthesizing data locally or perturbing individual records in federated settings produces biased mixtures or introduces excessive noise under heterogeneous data distributions.technical 85%
• HeteroFedSyn achieves utility comparable to centralized synthesis despite increased noise inherent to federated execution.result 88%
• HeteroFedSyn introduces three key innovations for distributed marginal selection: L2-based dependency metric with random projection, unbiased estimator to correct multiplicative noise, and adaptive selection strategy.technical 92%
• Traditional differential privacy mechanisms are typically tailored to specific analysis tasks, limiting reusability of protected data.limitation 90%
• DP tabular data synthesis overcomes task-specific limitations by generating synthetic datasets that can be shared for arbitrary downstream tasks.technical 85%
• HeteroFedSyn is built upon the PrivSyn paradigm of 2-way marginal-based synthesis.method 95%

Competitive landscape

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

Startup potential card