Timezone: »
We investigate the problem of learning discrete graphical models in a differentially private way. Approaches to this problem range from privileged algorithms that conduct learning completely behind the privacy barrier to schemes that release private summary statistics paired with algorithms to learn parameters from those statistics. 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.
Author Information
Garrett Bernstein (University of Massachusetts Amherst)
Ryan McKenna (UMass Amherst)
Tao Sun (University of Massachusetts Amherst)
Daniel Sheldon (University of Massachusetts Amherst)
Michael Hay (Colgate University)
Gerome Miklau (University of Massachusetts, Amherst)
Related Events (a corresponding poster, oral, or spotlight)
-
2017 Poster: Differentially Private Learning of Graphical Models using CGMs »
Wed. Aug 9th 08:30 AM -- 12:00 PM Room Gallery #51
More from the Same Authors
-
2021 : Differentially Private Classification via 0-1 Loss »
Ryan McKenna -
2021 : A Practitioners Guide to Differentially Private Convex Optimization »
Ryan McKenna · Hristo Paskov · Kunal Talwar -
2023 Poster: Automatically marginalized MCMC in probabilistic programming »
Jinlin Lai · Javier Burroni · Hui Guan · Daniel Sheldon -
2021 : Contributed Talks Session 2 »
Saeed Sharifi-Malvajerdi · Audra McMillan · Ryan McKenna -
2019 : AI for Ecology and Conservation »
Daniel Sheldon -
2019 Poster: Graphical-model based estimation and inference for differential privacy »
Ryan McKenna · Daniel Sheldon · Gerome Miklau -
2019 Oral: Graphical-model based estimation and inference for differential privacy »
Ryan McKenna · Daniel Sheldon · Gerome Miklau -
2018 Poster: Learning in Integer Latent Variable Models with Nested Automatic Differentiation »
Daniel Sheldon · Kevin Winner · Debora Sujono -
2018 Oral: Learning in Integer Latent Variable Models with Nested Automatic Differentiation »
Daniel Sheldon · Kevin Winner · Debora Sujono -
2017 Poster: Exact Inference for Integer Latent-Variable Models »
Kevin Winner · Debora Sujono · Daniel Sheldon -
2017 Talk: Exact Inference for Integer Latent-Variable Models »
Kevin Winner · Debora Sujono · Daniel Sheldon