The tutorial will introduce attendees to Kolmogorov-Arnold Networks (KANs), a new machine learning framework that promises greater interpretability and controllability than multilayer perceptrons (MLPs). In addition to defining KANs and explaining how they are trained, the tutorial will guide attendees (through concrete examples) in understanding how to interpret and manually control a trained KAN. It will also present the current state of the art in terms of proposed architectures within framework and their applications to various tasks. As KANs represent an innovative and potentially impactful advancement in AI, our goal is to help attendees thoroughly understand their advantages and limitations, to encourage the community to explore their use across different domains and tasks.
Description
The tutorial introduces attendees to Kolmogorov-Arnold Networks (KANs), a new approach to machine learning that was recently proposed as a more interpretable and controllable framework than the ubiquitous multi-layer perceptron. Besides providing an understanding of the KAN framework and its position within the overall machine-learning picture, the tutorial will also guide attendees through the different KAN(-based) architectures that have been proposed in the literature and their applications.
The target audience is machine learning researchers and practitioners, and especially those working on deep neural networks. Besides the more theoretical understanding of KAN architectures, the tutorial will also present examples of how the interpretability and controllability of KANs is concretely realized. Finally, the tutorial will discuss advantages and limitations of KANs for different application areas with the aim to motivate interested members of the audience to experiment with them.
About Us
Tatiana Boura (BSc in Informatics and Telecommunications, MSc in AI) is affiliated to the Institute of Informatics and Telecommunications, NCSR Demokritos working on cutting-edge AI-centered projects, focusing on the field of explainable AI (xAI), with an emphasis on neurosymbolic artificial intelligence. She is a member of the Greek AI Society.
Stasinos Konstantopoulos (MEng in Computer Engineering and Informatics, MSc in AI, PhD in ILP) is affiliated to the Institute of Informatics and Telecommunications, NCSR Demokritos where he has worked on various areas of AI from computational logics to robot sensing to machine learning. He is a member of the Greek AI Society and served on the society’s 2021-2022 Board. He is a regular reviewer for AI and Semantic Web conferences and journals, including ECAI 2025. He is also an instructor at the MSc on AI jointly organized by University of Piraeus and NCSR Demokritos.
Background
Tatiana and Stasinos’ research on KANs focuses on recurrency and sequence processing where they developed seqKAN, a new KAN architecture ( arxiv:2502.14681 )
In November 2024 they co-organized and jointly presented a 2-hour internal talk and workshop at the Institute of Informatics and Telecommunications, NCSR Demokritos with the aim to present the new framework to colleagues, present their own work on the subject, and motivate colleagues to apply KANs to different tasks.
Slides and bibliography handout are available at github:data-eng/KAN . The event attracted around 40 participants (cf. iit:talk-on-KANs ).
Contact
Let us get in touch!
Acknowledgements
This work was co-funded by the European Union under GA no. 101135782 MANOLO . Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or CNECT. Neither the European Union nor CNECT can be held responsible for them.
The cover image is from Pexels .