A Stopping Criterion for Transductive Active Learning

Daniel Kottke; Christoph Sandrock; Georg Krempl; Bernhard Sick

doi:10.1007/978-3-031-26412-2_29

A Stopping Criterion for Transductive Active Learning

Daniel Kottke^*, Christoph Sandrock, Georg Krempl, Bernhard Sick

^*Corresponding author for this work

University of Kassel

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

Abstract

In transductive active learning, the goal is to determine the correct labels for an unlabeled, known dataset. Therefore, we can either ask an oracle to provide the right label at some cost or use the prediction of a classifier which we train on the labels acquired so far. In contrast, the commonly used (inductive) active learning aims to select instances for labeling out of the unlabeled set to create a generalized classifier, which will be deployed on unknown data. This article formally defines the transductive setting and shows that it requires new solutions. Additionally, we formalize the theoretically cost-optimal stopping point for the transductive scenario. Building upon the probabilistic active learning framework, we propose a new transductive selection strategy that includes a stopping criterion and show its superiority.

Original language	English
Title of host publication	Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings
Editors	Massih-Reza Amini, Stéphane Canu, Asja Fischer, Tias Guns, Petra Kralj Novak, Grigorios Tsoumakas
Publisher	Springer
Pages	468-484
Number of pages	17
ISBN (Print)	9783031264115
DOIs	https://doi.org/10.1007/978-3-031-26412-2_29
Publication status	Published - 17 Mar 2023
Event	22nd Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2022 - Grenoble, France Duration: 19 Sept 2022 → 23 Sept 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13716 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	22nd Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2022
Country/Territory	France
City	Grenoble
Period	19/09/22 → 23/09/22

Bibliographical note

Publisher Copyright:
© 2023, The Author(s).

Keywords

Active learning
Stopping criteria
Transduction

Access to Document

10.1007/978-3-031-26412-2_29Licence: CC BY

978-3-031-26412-2_29Final published version, 1.2 MBLicence: CC BY

Cite this

Kottke, D., Sandrock, C., Krempl, G., & Sick, B. (2023). A Stopping Criterion for Transductive Active Learning. In M.-R. Amini, S. Canu, A. Fischer, T. Guns, P. Kralj Novak, & G. Tsoumakas (Eds.), Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings (pp. 468-484). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13716 LNAI). Springer. https://doi.org/10.1007/978-3-031-26412-2_29

Kottke, Daniel ; Sandrock, Christoph ; Krempl, Georg et al. / A Stopping Criterion for Transductive Active Learning. Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings. editor / Massih-Reza Amini ; Stéphane Canu ; Asja Fischer ; Tias Guns ; Petra Kralj Novak ; Grigorios Tsoumakas. Springer, 2023. pp. 468-484 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{14c2a8e81027472793cda2bebeb67004,

title = "A Stopping Criterion for Transductive Active Learning",

abstract = "In transductive active learning, the goal is to determine the correct labels for an unlabeled, known dataset. Therefore, we can either ask an oracle to provide the right label at some cost or use the prediction of a classifier which we train on the labels acquired so far. In contrast, the commonly used (inductive) active learning aims to select instances for labeling out of the unlabeled set to create a generalized classifier, which will be deployed on unknown data. This article formally defines the transductive setting and shows that it requires new solutions. Additionally, we formalize the theoretically cost-optimal stopping point for the transductive scenario. Building upon the probabilistic active learning framework, we propose a new transductive selection strategy that includes a stopping criterion and show its superiority.",

keywords = "Active learning, Stopping criteria, Transduction",

author = "Daniel Kottke and Christoph Sandrock and Georg Krempl and Bernhard Sick",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s).; 22nd Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2022 ; Conference date: 19-09-2022 Through 23-09-2022",

year = "2023",

month = mar,

day = "17",

doi = "10.1007/978-3-031-26412-2_29",

language = "English",

isbn = "9783031264115",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "468--484",

editor = "Massih-Reza Amini and St{\'e}phane Canu and Asja Fischer and Tias Guns and {Kralj Novak}, Petra and Grigorios Tsoumakas",

booktitle = "Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings",

}

Kottke, D, Sandrock, C, Krempl, G & Sick, B 2023, A Stopping Criterion for Transductive Active Learning. in M-R Amini, S Canu, A Fischer, T Guns, P Kralj Novak & G Tsoumakas (eds), Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13716 LNAI, Springer, pp. 468-484, 22nd Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2022, Grenoble, France, 19/09/22. https://doi.org/10.1007/978-3-031-26412-2_29

A Stopping Criterion for Transductive Active Learning. / Kottke, Daniel; Sandrock, Christoph; Krempl, Georg et al.
Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings. ed. / Massih-Reza Amini; Stéphane Canu; Asja Fischer; Tias Guns; Petra Kralj Novak; Grigorios Tsoumakas. Springer, 2023. p. 468-484 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13716 LNAI).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - A Stopping Criterion for Transductive Active Learning

AU - Kottke, Daniel

AU - Sandrock, Christoph

AU - Krempl, Georg

AU - Sick, Bernhard

PY - 2023/3/17

Y1 - 2023/3/17

N2 - In transductive active learning, the goal is to determine the correct labels for an unlabeled, known dataset. Therefore, we can either ask an oracle to provide the right label at some cost or use the prediction of a classifier which we train on the labels acquired so far. In contrast, the commonly used (inductive) active learning aims to select instances for labeling out of the unlabeled set to create a generalized classifier, which will be deployed on unknown data. This article formally defines the transductive setting and shows that it requires new solutions. Additionally, we formalize the theoretically cost-optimal stopping point for the transductive scenario. Building upon the probabilistic active learning framework, we propose a new transductive selection strategy that includes a stopping criterion and show its superiority.

AB - In transductive active learning, the goal is to determine the correct labels for an unlabeled, known dataset. Therefore, we can either ask an oracle to provide the right label at some cost or use the prediction of a classifier which we train on the labels acquired so far. In contrast, the commonly used (inductive) active learning aims to select instances for labeling out of the unlabeled set to create a generalized classifier, which will be deployed on unknown data. This article formally defines the transductive setting and shows that it requires new solutions. Additionally, we formalize the theoretically cost-optimal stopping point for the transductive scenario. Building upon the probabilistic active learning framework, we propose a new transductive selection strategy that includes a stopping criterion and show its superiority.

KW - Active learning

KW - Stopping criteria

KW - Transduction

UR - http://www.scopus.com/inward/record.url?scp=85151061548&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-26412-2_29

DO - 10.1007/978-3-031-26412-2_29

M3 - Conference contribution

AN - SCOPUS:85151061548

SN - 9783031264115

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 468

EP - 484

BT - Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings

A2 - Amini, Massih-Reza

A2 - Canu, Stéphane

A2 - Fischer, Asja

A2 - Guns, Tias

A2 - Kralj Novak, Petra

A2 - Tsoumakas, Grigorios

PB - Springer

T2 - 22nd Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2022

Y2 - 19 September 2022 through 23 September 2022

ER -

Kottke D, Sandrock C, Krempl G, Sick B. A Stopping Criterion for Transductive Active Learning. In Amini MR, Canu S, Fischer A, Guns T, Kralj Novak P, Tsoumakas G, editors, Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2022, Proceedings. Springer. 2023. p. 468-484. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-26412-2_29