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: Gradient Descent on Infinitely Wide Neural Networks: Global Convergence and Generalization

Authors: Francis Bach (SIERRA), Lenaïc Chizat (EPFL)
Abstract: Many supervised machine learning methods are naturally cast as optimization problems. For prediction models which are linear in their parameters, this often leads to convex problems for which many mathematical guarantees exist. Models which are non-linear in their parameters such as neural networks lead to non-convex optimization problems for which guarantees are harder to obtain. In this review paper, we consider two-layer neural networks with homogeneous activation functions where the number of hidden neurons tends to infinity, and show how qualitative convergence guarantees may be derived.
Subjects: Machine Learning (cs.LG); Optimization and Control (math.OC); Statistics Theory (math.ST)
Cite as: arXiv:2110.08084 [cs.LG]
  (or arXiv:2110.08084v1 [cs.LG] for this version)

Submission history

From: Francis Bach [view email]
[v1] Fri, 15 Oct 2021 13:25:32 GMT (1952kb,D)

Link back to: arXiv, form interface, contact.