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

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

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

Computer Science > Machine Learning

Title: Adaptive Differentially Private Empirical Risk Minimization

Abstract: We propose an adaptive (stochastic) gradient perturbation method for differentially private empirical risk minimization. At each iteration, the random noise added to the gradient is optimally adapted to the stepsize; we name this process adaptive differentially private (ADP) learning. Given the same privacy budget, we prove that the ADP method considerably improves the utility guarantee compared to the standard differentially private method in which vanilla random noise is added. Our method is particularly useful for gradient-based algorithms with time-varying learning rates, including variants of AdaGrad (Duchi et al., 2011). We provide extensive numerical experiments to demonstrate the effectiveness of the proposed adaptive differentially private algorithm.
Subjects: Machine Learning (cs.LG); Image and Video Processing (eess.IV); Optimization and Control (math.OC); Machine Learning (stat.ML)
Cite as: arXiv:2110.07435 [cs.LG]
  (or arXiv:2110.07435v2 [cs.LG] for this version)

Submission history

From: Xiaoixa Wu [view email]
[v1] Thu, 14 Oct 2021 15:02:20 GMT (4353kb,D)
[v2] Mon, 25 Oct 2021 01:35:54 GMT (4367kb,D)

Link back to: arXiv, form interface, contact.