Lookup multivariate Kolmogorov-Arnold Networks
Sergey Pozdnyakov [email protected], Philippe Schwaller [email protected], [-1.5em] École Polytechnique Fédérale de Lausanne (EPFL) · 2025
Read Lookup multivariate Kolmogorov-Arnold Networks by Sergey Pozdnyakov [email protected], Philippe Schwaller [email protected], [-1.5em] École Polytechnique Fédérale de Lausanne (EPFL) (2025) 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-2509-07103.txt.
Description
Sergey Pozdnyakov and Philippe Schwaller introduce lookup multivariate Kolmogorov-Arnold Networks in a 2025 paper. In most modern deep learning models, the big linear layers eat up most of the parameters and most of the computing cost.
Their design builds a large mapping out of many small ones. Each small function works in few dimensions, yet can hold dozens or hundreds of trainable parameters. Because each one is stored as a spline lookup table, working it out takes only a few multiplications. In tests, the networks cut the arithmetic needed at inference time by a large factor. At the same time they matched multilayer perceptrons at fitting general functions. In a second test they gave much higher throughput for the same accuracy. The authors also put the idea to work inside convolutional systems.
Catalog record
| CREATOR | Sergey Pozdnyakov [email protected], Philippe Schwaller [email protected], [-1.5em] École Polytechnique Fédérale de Lausanne (EPFL) |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 85 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-07103