Activity Number:
|
268
- A Unifying Theme for Interpretable Information Extraction from Data: The Stability Principle
|
Type:
|
Invited
|
Date/Time:
|
Tuesday, August 1, 2017 : 8:30 AM to 10:20 AM
|
Sponsor:
|
IMS
|
Abstract #322076
|
View Presentation
|
Title:
|
Max-Information, Differential Privacy, and Post-Selection Hypothesis Testing
|
Author(s):
|
Ryan Rogers and Aaron Roth* and Adam Smith and Om Thakkar
|
Companies:
|
University of Pennsylvania and University of Pennsylvania and Pennsylvania State University and Penn State
|
Keywords:
|
Selective Inference ;
Post Selection Inference ;
Hypothesis Testing ;
Differential Privacy ;
Adaptive Data Analysis
|
Abstract:
|
We study how the generalization properties of differential privacy can be used to perform adaptive hypothesis testing, while giving statistically valid p-value corrections. We do this by observing that the guarantees of algorithms with bounded "approximate max-information" are sufficient to correct the p-values of hypotheses which have been chosen with arbitrary degrees of adaptivity, and then by proving that algorithms that satisfy (epsilon,delta)-differential privacy have bounded approximate max information. The talk will be aimed at a general statistical audience who does not have background in differential privacy.
|
Authors who are presenting talks have a * after their name.