Spectral Gating Networks
Jusheng Zhang, Yijia Fan, Kaitong Cai, Jing Yang, Yongsen Zheng, Kwok-Yan Lam, Liang Lin, Keze Wang · 2026
Read Spectral Gating Networks by Jusheng Zhang, Yijia Fan, Kaitong Cai, Jing Yang, Yongsen Zheng, Kwok-Yan Lam, Liang Lin, Keze Wang (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-2602-07679.txt.
Description
Feed-forward networks need frequency-rich expressivity without losing stability, a tension exposed by spline-based Kolmogorov-Arnold networks whose grid refinement bloats weight counts and brittles tuning. Jusheng Zhang and seven co-authors present Spectral Gating Networks, a drop-in spectral reparameterization for standard MLP layers under fixed setting and training budgets.
A compact spectral pathway built from trainable random Fourier features joins the standard activation pathway. Learnable gates that start from stable base action and progressively allocate capacity to spectral features during training. Network builders get spectral expressivity that grows safely, rather than a parameterization that must be tuned to stay stable.
Catalog record
| CREATOR | Jusheng Zhang, Yijia Fan, Kaitong Cai, Jing Yang, Yongsen Zheng, Kwok-Yan Lam, Liang Lin, Keze Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 91 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07679