A Graph Meta-Network for Learning on Kolmogorov-Arnold Networks
Guy Bar-Shalom, Ami Tavory, Itay Evron, Maya Bechler-Speicher, Ido Guy, Haggai Maron · 2026
Read A Graph Meta-Network for Learning on Kolmogorov-Arnold Networks by Guy Bar-Shalom, Ami Tavory, Itay Evron, Maya Bechler-Speicher, Ido Guy, Haggai Maron (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-16316.txt.
Description
Guy Bar-Shalom and coauthors develop a graph meta-network for learning on Kolmogorov-Arnold Networks. Weight-space models learn directly from neural network parameters, enabling tasks such as predicting accuracy on new data sets. Naive methods perform poorly, making better weight-space architectures a central challenge. Prior work leveraged permutation rules in standard networks to guide designs, but no analogous analysis or tailored architecture yet existed for KANs.
The authors show that KANs share the same permutation rules as multilayer perceptrons, and propose the KAN-graph, a graph representation of their computation. Building on that, they develop WS-KAN, the first weight-space architecture that learns on KANs, accounting for their symmetry. They show WS-KAN can replicate an input network forward pass. A zoo of trained networks spanning diverse tasks serves as benchmark, and WS-KAN consistently outperforms structure-agnostic baselines.
Catalog record
| CREATOR | Guy Bar-Shalom, Ami Tavory, Itay Evron, Maya Bechler-Speicher, Ido Guy, Haggai Maron |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-16316