[Ml-stat-talks] Thurs: John Duchi on privacy-preserving machine learning

David Mimno mimno at CS.Princeton.EDU
Mon Apr 1 09:53:24 EDT 2013


There's a hitch in the big data revolution: a lot of that information is
private, sensitive, or distributed. This week we welcome John Duchi from
Berkeley, who will discuss statistical methods that do not require access
to the original data.

Thurs Apr 4, CS402, 12:30

John Duchi, UC Berkeley
Local privacy, minimax rates, and learning

Abstract: Working under a model of privacy in which data remains private
even from the statistician, we study the tradeoff between privacy
guarantees and the utility of the resulting statistical estimators. We
prove bounds on information-theoretic quantities, including mutual
information and Kullback-Leibler divergence, that influence estimation
rates as a function of the amount of privacy preserved. When combined with
standard minimax techniques such as Le Cam's and Fano's methods, these
inequalities allow for a precise characterization of statistical rates
under local privacy constraints. In this paper, we provide a complete
treatment of three canonical problem families: mean estimation in location
family models, parameter estimation in fixed-design regression, and convex
risk minimization. For all of these families, we provide lower and upper
bounds that match up to constant factors, giving privacy-preserving
mechanisms and computationally efficient estimators that achieve the bounds.

Bio: John Duchi is a fifth year PhD candidate in computer science at
Berkeley, working in the Statistical Artificial Intelligence Lab (SAIL)
under the joint supervision of Michael Jordan and Martin Wainwright. He has
master's degrees in statistics from Berkeley and computer science from
Stanford, where he worked with Daphne Koller. John has also worked in the
research team at Google with Yoram Singer. John's research has won and been
supported by several awards, including a best paper award at the
International Conference on Machine Learning (ICML) and NDSEG and Facebook
graduate fellowships.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20130401/396fee5d/attachment.html>


More information about the Ml-stat-talks mailing list