Optimistic semi-supervised least squares classification

Jesse H. Krijthe; Marco Loog

doi:10.1109/ICPR.2016.7899877

Optimistic semi-supervised least squares classification

Jesse H. Krijthe, Marco Loog

Datalogisk Institut

4 Citationer (Scopus)

Abstract

The goal of semi-supervised learning is to improve supervised classifiers by using additional unlabeled training examples. In this work we study a simple self-learning approach to semi-supervised learning applied to the least squares classifier. We show that a soft-label and a hard-label variant of self-learning can be derived by applying block coordinate descent to two related but slightly different objective functions. The resulting soft-label approach is related to an idea about dealing with missing data that dates back to the 1930s. We show that the soft-label variant typically outperforms the hard-label variant on benchmark datasets and partially explain this behaviour by studying the relative difficulty of finding good local minima for the corresponding objective functions.

Originalsprog	Engelsk
Titel	23rd International Conference on Pattern Recognition, ICPR 2016
Antal sider	6
Forlag	IEEE
Publikationsdato	1 jan. 2016
Sider	1677-1682
ISBN (Elektronisk)	978-1-5090-4847-2
DOI	https://doi.org/10.1109/ICPR.2016.7899877
Status	Udgivet - 1 jan. 2016
Begivenhed	23rd International Conference on Pattern Recognition - Cancun, Mexico Varighed: 4 dec. 2016 → 8 dec. 2016 Konferencens nummer: 23

Konference

Konference	23rd International Conference on Pattern Recognition
Nummer	23
Land/Område	Mexico
By	Cancun
Periode	04/12/2016 → 08/12/2016

Adgang til dokumentet

10.1109/ICPR.2016.7899877

Andre filer og links

Link to publication in Scopus

Citationsformater

@inproceedings{ab9a0b187f3d4e69bf08f6eb1e4054ae,

title = "Optimistic semi-supervised least squares classification",

abstract = "The goal of semi-supervised learning is to improve supervised classifiers by using additional unlabeled training examples. In this work we study a simple self-learning approach to semi-supervised learning applied to the least squares classifier. We show that a soft-label and a hard-label variant of self-learning can be derived by applying block coordinate descent to two related but slightly different objective functions. The resulting soft-label approach is related to an idea about dealing with missing data that dates back to the 1930s. We show that the soft-label variant typically outperforms the hard-label variant on benchmark datasets and partially explain this behaviour by studying the relative difficulty of finding good local minima for the corresponding objective functions.",

author = "Krijthe, {Jesse H.} and Marco Loog",

year = "2016",

month = jan,

day = "1",

doi = "10.1109/ICPR.2016.7899877",

language = "English",

pages = "1677--1682",

booktitle = "23rd International Conference on Pattern Recognition, ICPR 2016",

publisher = "IEEE",

note = "23rd International Conference on Pattern Recognition ; Conference date: 04-12-2016 Through 08-12-2016",

}

TY - GEN

T1 - Optimistic semi-supervised least squares classification

AU - Krijthe, Jesse H.

AU - Loog, Marco

N1 - Conference code: 23

PY - 2016/1/1

Y1 - 2016/1/1

N2 - The goal of semi-supervised learning is to improve supervised classifiers by using additional unlabeled training examples. In this work we study a simple self-learning approach to semi-supervised learning applied to the least squares classifier. We show that a soft-label and a hard-label variant of self-learning can be derived by applying block coordinate descent to two related but slightly different objective functions. The resulting soft-label approach is related to an idea about dealing with missing data that dates back to the 1930s. We show that the soft-label variant typically outperforms the hard-label variant on benchmark datasets and partially explain this behaviour by studying the relative difficulty of finding good local minima for the corresponding objective functions.

AB - The goal of semi-supervised learning is to improve supervised classifiers by using additional unlabeled training examples. In this work we study a simple self-learning approach to semi-supervised learning applied to the least squares classifier. We show that a soft-label and a hard-label variant of self-learning can be derived by applying block coordinate descent to two related but slightly different objective functions. The resulting soft-label approach is related to an idea about dealing with missing data that dates back to the 1930s. We show that the soft-label variant typically outperforms the hard-label variant on benchmark datasets and partially explain this behaviour by studying the relative difficulty of finding good local minima for the corresponding objective functions.

UR - http://www.scopus.com/inward/record.url?scp=85018623397&partnerID=8YFLogxK

U2 - 10.1109/ICPR.2016.7899877

DO - 10.1109/ICPR.2016.7899877

M3 - Article in proceedings

AN - SCOPUS:85018623397

SP - 1677

EP - 1682

BT - 23rd International Conference on Pattern Recognition, ICPR 2016

PB - IEEE

T2 - 23rd International Conference on Pattern Recognition

Y2 - 4 December 2016 through 8 December 2016

ER -

Optimistic semi-supervised least squares classification

Abstract

Konference

Adgang til dokumentet

Andre filer og links

Fingeraftryk

Citationsformater