Local Rademacher Complexity: sharper risk bounds with and without unlabeled samples

Luca Oneto; Alessandro Ghio; Sandro Ridella; Davide Anguita

doi:10.1016/j.neunet.2015.02.006

Local Rademacher Complexity: sharper risk bounds with and without unlabeled samples

Neural Netw. 2015 May:65:115-25. doi: 10.1016/j.neunet.2015.02.006. Epub 2015 Feb 16.

Authors

Luca Oneto¹, Alessandro Ghio², Sandro Ridella³, Davide Anguita⁴

Affiliations

¹ DITEN - University of Genova, Via Opera Pia 11A, I-16145 Genova, Italy. Electronic address: Luca.Oneto@unige.it.
² DIBRIS - University of Genova, Via Opera Pia 13, I-16145 Genova, Italy. Electronic address: Alessandro.Ghio@unige.it.
³ DITEN - University of Genova, Via Opera Pia 11A, I-16145 Genova, Italy. Electronic address: Sandro.Ridella@unige.it.
⁴ DIBRIS - University of Genova, Via Opera Pia 13, I-16145 Genova, Italy. Electronic address: Davide.Anguita@unige.it.

PMID: 25734890
DOI: 10.1016/j.neunet.2015.02.006

Abstract

We derive in this paper a new Local Rademacher Complexity risk bound on the generalization ability of a model, which is able to take advantage of the availability of unlabeled samples. Moreover, this new bound improves state-of-the-art results even when no unlabeled samples are available.

Keywords: Local Rademacher Complexity; Performance estimation; Statistical learning theory; Unlabeled samples.

MeSH terms

Artificial Intelligence*
Models, Statistical*