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

Download:

Current browse context:

cs

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

Statistics > Machine Learning

Title: Stability and Generalization of the Decentralized Stochastic Gradient Descent

Abstract: The stability and generalization of stochastic gradient-based methods provide valuable insights into understanding the algorithmic performance of machine learning models. As the main workhorse for deep learning, stochastic gradient descent has received a considerable amount of studies. Nevertheless, the community paid little attention to its decentralized variants. In this paper, we provide a novel formulation of the decentralized stochastic gradient descent. Leveraging this formulation together with (non)convex optimization theory, we establish the first stability and generalization guarantees for the decentralized stochastic gradient descent. Our theoretical results are built on top of a few common and mild assumptions and reveal that the decentralization deteriorates the stability of SGD for the first time. We verify our theoretical findings by using a variety of decentralized settings and benchmark machine learning models.
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG); Optimization and Control (math.OC)
Journal reference: AAAI-2021
Cite as: arXiv:2102.01302 [stat.ML]
  (or arXiv:2102.01302v3 [stat.ML] for this version)

Submission history

From: Tao Sun [view email]
[v1] Tue, 2 Feb 2021 04:23:23 GMT (308kb)
[v2] Tue, 9 Feb 2021 12:40:12 GMT (308kb)
[v3] Tue, 23 Feb 2021 01:24:35 GMT (308kb)

Link back to: arXiv, form interface, contact.