Research Paper#Privacy-Preserving Machine Learning, Graph Analysis🔬 ResearchAnalyzed: Jan 4, 2026 00:15
Private Inference in Directed Networks
Analysis
This paper addresses the problem of releasing directed graphs while preserving privacy. It focuses on the $p_0$ model and uses edge-flipping mechanisms under local differential privacy. The core contribution is a private estimator for the model parameters, shown to be consistent and normally distributed. The paper also compares input and output perturbation methods and applies the method to a real-world network.
Key Takeaways
- •Proposes a method for private inference in directed networks using the $p_0$ model.
- •Employs edge-flipping mechanisms under local differential privacy.
- •Provides a private estimator with theoretical guarantees (consistency, normality).
- •Compares different perturbation methods for privacy protection and accuracy.
- •Applies the method to a real-world network (UC Irvine message network).
Reference
“The paper introduces a private estimator for the $p_0$ model parameters and demonstrates its asymptotic properties.”