2017/06/14 by Garrett Bernstein, Bernstein, Garrett, Ryan McKenna +12
Computer Science · Mathematics · #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Algorithms #Mobile Crowdsensing and Crowdsourcing #Privacy-Preserving Technologies in Data #Statistical Methods and Bayesian Inference #cs.CR #cs.LG #stat.ML
paper · pdf · doi:10.48550/arxiv.1706.04646
Accepted to ICML 2017
arxiv created 2017/06/14 · openalex publication_date 2017/06/14 · arxiv updated 2017/06/16 · openalex created_date 2022/10/06 · openalex updated_date 2026/07/28
We investigate the problem of learning discrete, undirected graphical models in a differentially private way. We show that the approach of releasing noisy sufficient statistics using the Laplace mechanism achieves a good trade-off between privacy, utility, and practicality. A naive learning algorithm that uses the noisy sufficient statistics "as is" outperforms general-purpose differentially private learning algorithms. However, it has three limitations: it ignores knowledge about the data generating process, rests on uncertain theoretical foundations, and exhibits certain pathologies. We develop a more principled approach that applies the formalism of collective graphical models to perform inference over the true sufficient statistics within an expectation-maximization framework. We show that this learns better models than competing approaches on both synthetic data and on real human mobility data used as a case study.