Generative System Dynamics in Recurrent Neural Networks 1 1 1 Copyright © 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional pu
Michele Casoni, Tommaso Guidi, Alessandro Betti, Stefano Melacci, Marco Gori (1) Department of Information Engineering and Mathematics, University of Siena, Siena, Italy(2) Department of Information Engineering, University of Florence, Florence, Italy(3) IMT School for Advanced Studies Lucca, Lucca, · 2025
Read Generative System Dynamics in Recurrent Neural Networks 1 1 1 Copyright © 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional pu by Michele Casoni, Tommaso Guidi, Alessandro Betti, Stefano Melacci, Marco Gori (1) Department of Information Engineering and Mathematics, University of Siena, Siena, Italy(2) Department of Information Engineering, University of Florence, Florence, Italy(3) IMT School for Advanced Studies Lucca, Lucca, (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-2504-13951.txt.
Description
Michele Casoni and co-authors study, in 2025, when recurrent neural networks keep oscillating instead of settling down. Such networks process sequences over time, and steady cycles rather than fixed points may help them remember. The team studies their continuous-time behavior with nonlinear activation functions.
Skew-symmetric weight matrices turn out to be the key ingredient for stable limit cycles, in both linear and nonlinear versions. S-shaped activation functions of a common kind preserve the cycles and add conserved quantities to the motion. Simulations also show the nonlinear versions integrate more stably. The findings guide designs for networks that must hold long temporal patterns.
Catalog record
| CREATOR | Michele Casoni, Tommaso Guidi, Alessandro Betti, Stefano Melacci, Marco Gori (1) Department of Information Engineering and Mathematics, University of Siena, Siena, Italy(2) Department of Information Engineering, University of Florence, Florence, Italy(3) IMT School for Advanced Studies Lucca, Lucca, |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-13951