Reproduce “Graphical-model based estimation and inference for differential privacy” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ryan McKenna, Daniel Sheldon, Gerome Miklau
Researchers from the University of Massachusetts, Amherst introduce a graphical model-based framework for differentially private estimation and inference, addressing scalability limitations in high-dimensional data while significantly enhancing accuracy. This approach allowed algorithms like MWEM and HDMM to run on datasets with up to 1000 attributes, previously infeasible, and reduced error by up to 27 times on various benchmarks.
Key insights
- Representing data distributions with graphical models under the maximum entropy principle allows for compact, scalable estimation in high-dimensional settings.
- Convex optimization, specifically proximal algorithms with a MARGINAL-ORACLE, provides an efficient and theoretically robust way to find optimal marginals for graphical models.
- The proposed graphical model-based estimation (PGM estimation) is modular, enhancing the accuracy and scalability of diverse differentially private mechanisms without additional privacy cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.