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: Potential-Function Proofs for First-Order Methods

Abstract: This note discusses proofs for convergence of first-order methods based on simple potential-function arguments. We cover methods like gradient descent (for both smooth and non-smooth settings), mirror descent, and some accelerated variants.
Subjects: Machine Learning (cs.LG); Data Structures and Algorithms (cs.DS); Optimization and Control (math.OC)
Cite as: arXiv:1712.04581 [cs.LG]
  (or arXiv:1712.04581v3 [cs.LG] for this version)

Submission history

From: Nikhil Bansal [view email]
[v1] Wed, 13 Dec 2017 01:10:13 GMT (171kb,D)
[v2] Thu, 7 Mar 2019 21:55:56 GMT (177kb,D)
[v3] Sun, 2 Jun 2019 19:36:25 GMT (177kb,D)

Link back to: arXiv, form interface, contact.