מושגי ליבה
Developing private algorithms for graphon estimation with polynomial running time.
תקציר
Introduction
Differential privacy in graph data.
Edge-differential vs. node-differential privacy.
Techniques
Score function and exponential mechanism.
Sum-of-squares relaxation for optimization problems.
Lipschitz extensions for privacy guarantees.
Data Extraction
"The algorithm is based on an exponential mechanism for a score function defined in terms of a sum-of-squares relaxation whose level depends on the number of blocks."
Results
Private algorithm for learning stochastic block models with polynomial running time.
Algorithm for graphon estimation with improved error convergence rates.
Utility Analysis
Utility guarantees of exponential mechanisms based on score functions.
Lower Bound Analysis
Sample complexity lower bound for private estimation of stochastic block model.
Improvement in Non-Private Setting
Improved error rates compared to existing algorithms in non-private settings.
Inquiry and Critical Thinking Questions
סטטיסטיקה
The algorithm is based on an exponential mechanism for a score function defined in terms of a sum-of-squares relaxation whose level depends on the number of blocks.