Home // IARIA Congress 2025, The 2025 IARIA Annual Congress on Frontiers in Science, Technology, Services, and Applications // View article

Progressively Overcoming Catastrophic Forgetting in Kolmogorov–Arnold Networks

Authors:
Evgenii Ostanin
Nebojsa Djosic
Fatima Hussain
Salah Sharieh
Alexander Ferworn
Malek Sharieh

Keywords: Continual Learning, Catastrophic Forgetting, Kolmogorov–Arnold Networks, KAN, Spline Freezing, Memory Re tention, Experience Replay, Progressive Freezing.

Abstract:
Catastrophic forgetting remains a major challenge in continuous learning, particularly for architectures not explicitly designed for knowledge retention. This paper explores Kolmogorov–Arnold networks as an alternative to multilaye perceptrons in such settings. We introduce two freezing strategies: tensor-level spline freezing and point-level control freezing, that exploit the spline-based structure of Kolmogorov–Arnold networks to preserve knowledge from earlier tasks. Experiments on Modified National Institute of Standards and Technology (MNIST) handwritten digit dataset show that both methods yield modest but consistent improvements when paired with replay techniques. The best configurations improve total accuracy by up to 2.2% and reduce forgetting by 5.4% over the no-freeze baseline. These findings reveal a new direction for mitigating forgetting through the selective control of spline parameters specific for the Kolmogorov-Arnold networks. Future work will explore a deeper integration with regularization and expansion methods to further enhance knowledge retention in continual learning.

Pages: 138 to 143

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-284-5

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025