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


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


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

Computer Science > Machine Learning

Title: Online Hyperparameter Search Interleaved with Proximal Parameter Updates

Abstract: There is a clear need for efficient algorithms to tune hyperparameters for statistical learning schemes, since the commonly applied search methods (such as grid search with N-fold cross-validation) are inefficient and/or approximate. Previously existing algorithms that efficiently search for hyperparameters relying on the smoothness of the cost function cannot be applied in problems such as Lasso regression.
In this contribution, we develop a hyperparameter optimization method that relies on the structure of proximal gradient methods and does not require a smooth cost function. Such a method is applied to Leave-one-out (LOO)-validated Lasso and Group Lasso to yield efficient, data-driven, hyperparameter optimization algorithms.
Numerical experiments corroborate the convergence of the proposed method to a local optimum of the LOO validation error curve, and the efficiency of its approximations.
Comments: 6 pages, 3 figures, 1 algorithm; Submitted to the European Signal Processing Conference (EUSIPCO) 2020 (Amsterdam)
Subjects: Machine Learning (cs.LG); Signal Processing (eess.SP); Machine Learning (stat.ML)
Cite as: arXiv:2004.02769 [cs.LG]
  (or arXiv:2004.02769v1 [cs.LG] for this version)

Submission history

From: Luis Miguel López-Ramos [view email]
[v1] Mon, 6 Apr 2020 15:54:03 GMT (519kb)

Link back to: arXiv, form interface, contact.