We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Calibrating Noise to Variance in Adaptive Data Analysis

Abstract: Datasets are often used multiple times and each successive analysis may depend on the outcome of previous analyses. Standard techniques for ensuring generalization and statistical validity do not account for this adaptive dependence. A recent line of work studies the challenges that arise from such adaptive data reuse by considering the problem of answering a sequence of "queries" about the data distribution where each query may depend arbitrarily on answers to previous queries.
The strongest results obtained for this problem rely on differential privacy -- a strong notion of algorithmic stability with the important property that it "composes" well when data is reused. However the notion is rather strict, as it requires stability under replacement of an arbitrary data element. The simplest algorithm is to add Gaussian (or Laplace) noise to distort the empirical answers. However, analysing this technique using differential privacy yields suboptimal accuracy guarantees when the queries have low variance. Here we propose a relaxed notion of stability that also composes adaptively. We demonstrate that a simple and natural algorithm based on adding noise scaled to the standard deviation of the query provides our notion of stability. This implies an algorithm that can answer statistical queries about the dataset with substantially improved accuracy guarantees for low-variance queries. The only previous approach that provides such accuracy guarantees is based on a more involved differentially private median-of-means algorithm and its analysis exploits stronger "group" stability of the algorithm.
Comments: Accepted for presentation at Conference on Learning Theory (COLT) 2018
Subjects: Machine Learning (cs.LG); Cryptography and Security (cs.CR); Data Structures and Algorithms (cs.DS); Information Theory (cs.IT)
Cite as: arXiv:1712.07196 [cs.LG]
  (or arXiv:1712.07196v2 [cs.LG] for this version)

Submission history

From: Thomas Steinke [view email]
[v1] Tue, 19 Dec 2017 20:42:39 GMT (24kb)
[v2] Mon, 11 Jun 2018 19:14:14 GMT (36kb)

Link back to: arXiv, form interface, contact.