A4 Refereed article in a conference publication
Work-in-Progress: Evaluating Feasibility of Band Matrix Solvers for Scaling up Extreme Learning Machine Method
Authors: Akusok, Anton; Björk, Kaj-Mikael; Lendasse, Amaury; Espinosa Leal, Leonardo
Editors: Auer, Michael E.; Langmann, Reinhard; May, Dominik; Morales, Manuel
Conference name: International Conference on Smart Technologies & Education
Publisher: Springer Nature Switzerland
Publication year: 2026
Journal: Lecture Notes in Networks and Systems
Book title : Smart Technologies for an All-Electric Society : Proceedings of the 22nd International Conference on Smart Technologies & Education (STE2025). Volume 2
Volume: 1662
First page : 347
Last page: 355
ISBN: 978-3-032-07318-1
eISBN: 978-3-032-07319-8
ISSN: 2367-3370
eISSN: 2367-3389
DOI: https://doi.org/10.1007/978-3-032-07319-8_31
Publication's open availability at the time of reporting: No Open Access
Publication channel's open availability : Partially Open Access publication channel
Web address : https://doi.org/10.1007/978-3-032-07319-8_31
This work presents the results of the potential of band linear system solvers for improving the scalability of the Extreme Learning Machine (ELM) method at large model sizes. The model is tested on the standard MNIST dataset with a range of solvers provided by the SciPy Python library. The results are analyzed taking into consideration the overall performance and the performance impact of band solvers across different matrix bandwidths, as well as the performance versus runtime analysis. The findings show potential in applying the proposed method to very large ELM models with narrow band matrices.
