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

Download:

Current browse context:

stat.ML

Change to browse by:

References & Citations

Bookmark

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

Statistics > Machine Learning

Title: From inexact optimization to learning via gradient concentration

Abstract: Optimization in machine learning typically deals with the minimization of empirical objectives defined by training data. However, the ultimate goal of learning is to minimize the error on future data (test error), for which the training data provides only partial information. In this view, the optimization problems that are practically feasible are based on inexact quantities that are stochastic in nature. In this paper, we show how probabilistic results, specifically gradient concentration, can be combined with results from inexact optimization to derive sharp test error guarantees. By considering unconstrained objectives we highlight the implicit regularization properties of optimization for learning.
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG); Optimization and Control (math.OC)
Cite as: arXiv:2106.05397 [stat.ML]
  (or arXiv:2106.05397v3 [stat.ML] for this version)

Submission history

From: Lorenzo Rosasco [view email]
[v1] Wed, 9 Jun 2021 21:23:29 GMT (52kb,D)
[v2] Thu, 24 Jun 2021 20:11:34 GMT (43kb,D)
[v3] Fri, 5 Nov 2021 13:14:33 GMT (50kb,D)

Link back to: arXiv, form interface, contact.