Locally Private Bayesian Inference for Count Models

on YouTube ↗48m

" summary

Proposes locally private Bayesian inference for Poisson count models using the two-sided geometric mechanism for integer noise, rewrites noisy counts as differences of Poissons (Skellam) and introduces a Bessel auxiliary to sample true counts inside MCMC, then runs Poisson factorization (topic and community models). Shows on synthetic, State of the Union and Enron data that this posterior-aware method recovers much sparser means, often beating the naive truncate-and-fit baseline and nearly matching non-private error; notes scalability work moving to variational inference for speed.

AI-written summary · about

" chapters (14)

1. 0:00Introduction
2. 0:40Project Background
3. 3:25Bayesian Persona Factorization
4. 7:15Poisson Distribution
5. 9:05Example
6. 9:45Motivation
7. 20:19Data
8. 20:57Geometric Distribution
9. 32:40Email
10. 33:19Experiments
11. 36:33Data Set
12. 37:43Community Detection
13. 44:56Noise
14. 45:26Faster Inference

" related talks

• 01h 04mNonparametric Bayesian Methods: Models, Algorithms, and Applications IIMichael Jordan · Foundations of Machine Learning Boot Camp 2017
• 01h 06mNonparametric Bayesian Methods: Models, Algorithms, and Applications ITamara Broderick · Foundations of Machine Learning Boot Camp 2017
• 01h 02mNonparametric Bayesian Methods: Models, Algorithms, and Applications IIIMichael Jordan · Foundations of Machine Learning Boot Camp 2017
• 01h 07mStatistical Inference and Privacy, Part ISesa Slavkovic · Data Privacy: Foundations and Applications Boot Camp
• 43mPufferfish Privacy Mechanisms for Correlated DataKamalika Chaudhuri · Uncertainty in Computation 2016
• 42mBayesian Non-parametric Models for Data Science using PyMC3Christopher Fonnesbeck · PyCon 2018
• 39mLocal Algorithms for Community DetectionElchanan Mossel · Approximate Counting, Markov Chains and Phase Transitions
• 25mScaling Bayesian ExperimentationRobert J. Neal, Paul Cho · Scale By The Bay 2018