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 > Cryptography and Security

Title: Practical and Private (Deep) Learning without Sampling or Shuffling

Abstract: We consider training models with differential privacy (DP) using mini-batch gradients. The existing state-of-the-art, Differentially Private Stochastic Gradient Descent (DP-SGD), requires privacy amplification by sampling or shuffling to obtain the best privacy/accuracy/computation trade-offs. Unfortunately, the precise requirements on exact sampling and shuffling can be hard to obtain in important practical scenarios, particularly federated learning (FL). We design and analyze a DP variant of Follow-The-Regularized-Leader (DP-FTRL) that compares favorably (both theoretically and empirically) to amplified DP-SGD, while allowing for much more flexible data access patterns. DP-FTRL does not use any form of privacy amplification.
The code is available at this https URL and this https URL .
Subjects: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
Cite as: arXiv:2103.00039 [cs.CR]
  (or arXiv:2103.00039v3 [cs.CR] for this version)

Submission history

From: Shuang Song [view email]
[v1] Fri, 26 Feb 2021 20:16:26 GMT (410kb,D)
[v2] Tue, 15 Jun 2021 23:45:09 GMT (18862kb,D)
[v3] Fri, 10 Dec 2021 05:30:36 GMT (7606kb,D)

Link back to: arXiv, form interface, contact.