LUT-KAN: Segment-wise LUT Quantization for Fast KAN Inference
Oleksandr Kuznetsov · 2026
Read LUT-KAN: Segment-wise LUT Quantization for Fast KAN Inference by Oleksandr Kuznetsov (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-2601-03332.txt.
Description
Oleksandr Kuznetsov makes Kolmogorov-Arnold Networks fast enough for everyday hardware. These networks replace plain weights with learnable one-dimensional functions, usually splines. That makes them easy to interpret, but slow. Each layer needs many spline evaluations. Standard quantization tools do not help, since they assume matrix multiplication. His method, LUT-KAN, compiles each edge function into a per-segment lookup table, with eight-bit quantization and linear interpolation.
The package comes with an explicit inference contract. It states the boundary conventions and the rules for out-of-bounds inputs, so results can be reproduced. For honest speed claims, spline and lookup evaluation are compared under the same backends. That separates gains from the representation itself from gains due to vectorization or just-in-time effects. The aim is to keep the accuracy of the original networks while removing the CPU bottleneck.
Catalog record
| CREATOR | Oleksandr Kuznetsov |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-03332