Is Multitask Learning Always Better?

Mattick A, Mayr M, Maier A, Christlein V (2022)

Publication Type: Conference contribution

Publication year: 2022

Journal

Lecture Notes in Computer Science Springer Verlag

Publisher: Springer Science and Business Media Deutschland GmbH

Book Volume: 13237 LNCS

Pages Range: 674-687

Conference Proceedings Title: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Event location: La Rochelle FR

ISBN: 9783031065545

DOI: 10.1007/978-3-031-06555-2_45

Abstract

Multitask learning has been a common technique for improving representations learned by artificial neural networks for decades. However, the actual effects and trade-offs are not much explored, especially in the context of document analysis. We demonstrate a simple and realistic scenario on real-world datasets that produces noticeably inferior results in a multitask learning setting than in a single-task setting. We hypothesize that slight data-manifold and task semantic shifts are sufficient to lead to adversarial competition of tasks inside networks and demonstrate this experimentally in two different multitask learning formulations.

Authors with CRIS profile

Martin Mayr Lehrstuhl für Informatik 5 (Mustererkennung) Andreas Maier Lehrstuhl für Informatik 5 (Mustererkennung) Vincent Christlein Lehrstuhl für Informatik 5 (Mustererkennung)

Related research project(s)

Kommunikation und Sprache im Reich. Die Nürnberger Briefbücher im 15. Jahrhundert: Automatische Handschriftenerkennung - historische und sprachwissenschaftliche Analyse. Oct. 1, 2019 - Sept. 30, 2022

How to cite

APA:

Mattick, A., Mayr, M., Maier, A., & Christlein, V. (2022). Is Multitask Learning Always Better? In Seiichi Uchida, Elisa Barney, Véronique Eglin (Eds.), Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (pp. 674-687). La Rochelle, FR: Springer Science and Business Media Deutschland GmbH.

MLA:

Mattick, Alexander, et al. "Is Multitask Learning Always Better?" Proceedings of the 15th IAPR International Workshop on Document Analysis Systems, DAS 2022, La Rochelle Ed. Seiichi Uchida, Elisa Barney, Véronique Eglin, Springer Science and Business Media Deutschland GmbH, 2022. 674-687.

BibTeX: Download